Modern high-performance computing is moving toward architectures built on massive parallelism, instead of depending on further improvements in individual transistor performance. With device miniaturization approaching fundamental physical limits and , integrating larger numbers of cores onto a single chip has become the dominant strategy for sustaining computational growth. Graphics processing units (GPUs), especially those designed for artificial intelligence workloads, are a prime example. However, thermal management is arising as a real limitation and this is because the concentration of power dissipation within densely packed functional units leads to spatially heterogeneous temperature fields, where localized hot spots emerge and evolve dynamically in response to workload fluctuations. These thermal gradients directly influence device reliability, operational stability, and long-term degradation. As cooling strategies approach practical limits, sustaining performance increasingly depends on predictive models that can resolve fine spatial and temporal temperature variations across large-scale chips. Classical numerical approaches, including finite element, finite volume, and finite difference methods, remain the reference standard for thermal analysis due to their physical fidelity. However, their computational burden scales poorly with system size and resolution, which make them not suitable for real-time applications. Other methods were tried to reduce the cost by simplifying the physics or introducing surrogate representations such as circuit-based thermal models which can achieve speed through coarse spatial aggregation but may miss intra-unit hot spots and can approximate heat transfer pathways in a coarse way. Also, purely data-driven methods, such as neural networks, can provide efficiency, but they lack enforcement of the underlying equations, which can reduce confidence in predictions outside the training regime and make interpretation less direct.

Physics-based reduced-order modeling is currently emerging as a promising middle ground with proper orthogonal decomposition (POD), combined with Galerkin projection (GP), providing a structured way to reduce dimension but in the same time retain the underlying physics of heat transfer. Earlier implementations have demonstrated that such models can achieve high accuracy with a small number of modes when applied to systems with relatively few cores. The difficulty arises when extending these methods to architectures with thousands of interacting heat sources. Training such models becomes prohibitively expensive, as the number of possible power configurations grows rapidly. The central challenge, then, lies in preserving the efficiency and accuracy of physics-enforced reduced-order models while making them scalable to the size and complexity of modern many-core GPUs. In a recent research paper published in Structural and Multidisciplinary Optimization, Professor Lin Jiang from the College of Information Science and Engineering at Northeastern University  in China and Professors Yu Liu & Ming-Cheng Cheng from the Department of Electrical and Computer Engineering at Clarkson University developed a local ensemble POD-GP thermal modeling method that combines domain truncation with reusable generic sub-models.  The new approach is built around a different treatment of training locality. The underlying strategy begins by reconsidering how training data are generated and how the domain of the problem is represented. Instead of constructing a single global model or even assembling multiple full-domain models for individual power sources, the approach partitions the chip into a set of power source blocks, each representing one or several functional units. For each block, the thermal response is characterized independently. The key conceptual shift is that thermal influence from a localized source attenuates with distance, so the full chip does not need to be included in every training problem. By exploiting this behavior, the researchers truncate the computational domain around each power source block, restricting training to a localized region where temperature variations remain significant. This choice directly reduces the number of spatial degrees of freedom required in the simulations used to generate training data.

The authors estimated thermal length scales by analyzing how temperature decays away from a heat source, and the domain is extended to several multiples of this length to balance accuracy and efficiency. The analysis shows that extending the domain to approximately five thermal lengths captures the majority of the thermal contribution while avoiding unnecessary computational expense. This decision reflects a clear connection between physical behavior and model construction: the spatial extent of the training domain is dictated by heat diffusion characteristics rather than by geometric convenience. Once the research team established localized training domains, they applied POD to temperature fields generated from high-fidelity finite element simulations. The resulting POD modes provide a compact representation of the dominant thermal patterns within each truncated region. They used Galerkin projection to map the heat transfer equations onto this reduced space, producing a set of ordinary differential equations that describe the temporal evolution of modal coefficients. This step ensures that the reduced-order model remains anchored in the physics of heat conduction, avoiding the extrapolation issues associated with purely data-driven methods.

Many functional units within the GPU share identical or nearly identical geometries. Rather than training a separate model for each such unit, the method introduces generic truncated domains. A single trained model can therefore represent multiple identical regions, substantially reducing the total number of models required. For units located less than five thermal lengths from a chip boundary, a distinct truncated domain is trained separately to capture the boundary‑condition variations encoded in the finite‑element–generated temperature fields. In the case examined, a system comprising hundreds of power source blocks is represented by only a small set of generic models, each reused across multiple locations with appropriate spatial mapping. The assembled model reconstructs the global temperature field by superposing the contributions from all localized models. Because the underlying heat transfer process is linear within the operating temperature range considered, this superposition remains valid. The approach therefore combines localized accuracy with global coverage, without incurring the cost of full-domain simulations for every configuration. When applied to a large GPU architecture with more than ten thousand cores, the method produces detailed spatiotemporal temperature predictions that closely match those obtained from full finite element simulations. At the same time, the computational cost is reduced by several orders of magnitude. The model captures the emergence and evolution of dynamic hot spots across the chip, resolving fine spatial features that would be inaccessible to coarser modeling approaches.

The new approach of Professors Jiang, Liu & Cheng reconciles accuracy and computational efficiency by making locality part of the model architecture, rather than treating it as an after-the-fact simplification.  This distinguishes the method from purely data-driven surrogates, particularly in situations where operating conditions deviate from those seen during training. The ability to estimate error based on the eigenvalue spectrum of the POD modes introduces a level of predictability that is often absent in alternative approaches. It provides a quantitative link between model complexity and expected accuracy, which can guide practical implementation. The treatment of repeated structures within the chip further reflects a pragmatic understanding of modern hardware design. Many-core GPUs are characterized by a high degree of structural regularity, and the method turns this repetition into an opportunity for model reuse. The resulting approach enables detailed thermal analysis at scales that would otherwise be inaccessible for real-time or near-real-time applications. This has implications for dynamic thermal management strategies, where rapid prediction of temperature distributions is essential for controlling power and performance. Also, while the study focused on a specific GPU architecture, the underlying ideas could, in principle, extend to other large-scale integrated systems.

FIGURE LEGEND: Temperature maps of the Tesla Volta GV100 GPU at an instant in time calculated using (a) the finite element (FEniCS‑FEM) method and (b) the local ensemble POD‑GP (LEnPOD‑GP) model with six modes per truncated domain. Temperature profiles along the (c) x and (d) y directions through the high‑peak temperatures predicted by FEniCS‑FEM and LEn‑POD‑GP, along with the deviation of LEn‑POD‑GP from FEniCS‑FEM.  CREDIT: Struct Multidisc Optim 68, 235 (2025). https://doi.org/10.1007/s00158-025-04165-x

Optical flat metrology requires absolute surface evaluation when reference standards of higher accuracy are not available. Among the available approaches, the three-flat test has long been valued because it can recover the shapes of three unknown flats through self-consistent interferometric measurements. Extending that logic from line-based assessment to full-aperture surface reconstruction, however, has remained technically demanding, especially when high spatial fidelity must be retained across the reconstructed surface. Earlier full-field approaches generally followed two computational routes. Pixel-based methods reconstructed the surface directly from measured phase maps, which preserved finer spatial structure but introduced symmetry-linked residual components under rotational processing. Zernike-based methods instead expanded the measured data into modal coefficients, which gave greater flexibility in angle selection but removed higher spatial content once the retained order was fixed. That unresolved tension became more significant as interferometric repeatability improved. When hardware operates at sub-nanometer precision, the reconstruction model itself becomes the point at which that precision is either preserved or discarded. In a recent research paper published in Optics Express, Keita Tomita of Olympus Corporation, who is also a Ph.D. student, together with Dr. Toshiki Kumagai from Olympus Corporation, Dr. Kenichi Hibino from the National Institute of Advanced Industrial Science and Technology, and Dr. Satoru Takahashi from the University of Tokyo, addressed this problem by developing a symmetric full-field absolute testing algorithm for three optical flats based on three classical measurements and one additional rotated measurement. The method reconstructs each surface from a linear combination of 8N datasets created by systematic rotation and inversion of the measured phase maps. Its technical novelty lies in deriving the dataset weights from symmetry-based assignment equations and selecting the final coefficients through a minimum-norm Moore–Penrose pseudoinverse solution. That combination gives a pixel-based reconstruction whose first structured cross-talk term moves upward with N and whose spatial resolution improves as N increases.

The research team used four interferometric configurations: A with C, B with C, B with A, and B with a rotated C. In the rotated configuration, flat C turns by α = 2π/N, and each measured phase map is also paired with its inverted form. Repeated rotations of those maps generate 8N transformed datasets, and the reconstructed surface is written as a linear combination of all of them. That construction matters because it makes the reconstruction problem global at the dataset level. Instead of asking one measurement to carry a privileged share of the surface information, the method distributes the burden across a symmetry-complete family of rotated and inverted data.

The weight determination is central to the method. The authors first derive necessary assignment relations by inserting delta-function surfaces for A, B, or C and comparing where peaks must appear after transformation. Those conditions do not close exactly into a fully consistent system, so the authors introduce small constants o_j and rewrites the constraints in a modified but solvable form. The important point is not only algebraic convenience. Once the exact system becomes inconsistent, the reconstruction must be organized around an approximation principle that still respects the geometry of the problem. The authors choose uniform small constants tied to 1/6N and then determine the 8N weights by a minimum-norm solution obtained through the Moore–Penrose pseudoinverse. That decision gives the algorithm a controlled regularity: the weights are not chosen ad hoc, and the residual structure can be traced back to the same symmetric formulation that created the dataset family in the first place.

Their theoretical analysis then identifies the form of the remaining structured error. For Zernike components containing sin(Nθ), the reconstructed shape carries a cross-talk term equal to minus one third of the combined contribution from A, B, and C; for axial symmetry and cos(mθ) terms, the reconstruction error vanishes within the formulation they derive. This is important because it replaces a vague statement about “noise” with a very specific symmetry rule. The algorithm does not blur everything equally. It preserves some modal content exactly, and it pushes the first structured cross-talk to the N² + 1st Zernike component. Raising N is effective for a simple reason: the first symmetry-governed contamination is moved to higher order.  The simulations use sinusoidal grating surfaces rather than random Zernike combinations, which is a thoughtful choice because the paper has already shown that some modal classes reconstruct exactly and would mask trends if mixed indiscriminately. With N = 16, reconstruction at f = 3 lines per aperture showed no cross-talk error, and higher-frequency degradation emerged at f = 4 and 5. The interpolation error from bilinear rotation stayed below 0.005 nm rms in the case examined. The modulation transfer analysis then put numerical weight behind the design logic: the 99.9% MTF cut-off rose with N, reaching 4.78 lines per aperture for N = 16. At that same cut-off, the rms error stayed below 0.5 nm, and the investigators report that the N = 16 symmetric method reduced rms error to about one third of the N-position averaging result with the same nominal 8N degrees of freedom.

The laboratory validation was designed to test the same reconstruction logic. Using a wavelength-tuning Fizeau interferometer at 633 nm, with 15 phase-shifted images and a motorized rotation stage accurate to better than 0.05°, the authors reconstructed the absolute shape of a 60 mm clear-aperture flat for N = 2, 4, 8, and 16. Spatial detail increased as N increased. Cross-sectional comparison with the classical three-flat line result along the x-axis kept the deviations below 1 nm, and the rms difference dropped monotonically from 0.420 nm at N = 2 to 0.187 nm at N = 16. Repeated measurements of the three flats gave sub-nanometer standard deviations, listed as 0.697, 0.574, and 0.682 nm rms.

The new study by Keita Tomita and colleagues produced another reconstruction algorithm for optical flats and reorganized absolute three-flat testing around a tunable symmetry parameter and then shows, analytically and experimentally, what that parameter actually does to spatial fidelity. In a field where full-field reconstruction often becomes a negotiation among residual symmetry terms, modal truncation, and computational convenience, the paper supplies a cleaner design rule. N is no longer just a descriptive count associated with a special rotation. It becomes the knob that sets where structured cross-talk first appears in the modal hierarchy. That changes how one thinks about algorithm selection, because the trade is stated in a form that is visible before any measurement is taken.

There is also a methodological shift here. The paper keeps the reconstruction in pixel space, where fine spatial content can remain visible, yet it does not treat that choice as just empirical. The authors derive the weight equations from transformed delta responses, regularize the inconsistent constraint system in a controlled way, and then solve for minimum-norm coefficients through the pseudoinverse. That chain of reasoning matters and means the algorithm is both a better-performing recipe and also a reconstruction framework whose behavior can be tracked from symmetry assumptions to modal cross-talk, from modal cross-talk to transfer characteristics, and from transfer characteristics to the measured line-profile agreement with the classical test. An explicit analytical expression for the residual structure makes the reconstruction behavior easier to interpret.

The comparison with Zernike decomposition sharpens the broader meaning. The paper does not frame pixel-space and Zernike-space methods as interchangeable computational preferences. It shows that they preserve different kinds of information. Zernike reconstruction can remain clean when the retained order is chosen carefully, which is useful for low-spatial-frequency form evaluation. The symmetric pixel-based route keeps much finer spatial detail, because it does not discard the higher-order content through modal truncation. That distinction has practical weight in optical metrology. Production engineers interested in smooth low-order form may favor one representation; users who need to inspect finer surface structure have a strong reason to keep reconstruction in pixel space. The paper articulates that split in a way that is grounded in the measured and simulated behavior of the algorithms rather than in generic preference for one basis over another.

Another important implication comes from data use. The study notes that the N-position averaging method has the same nominal 8N data count as the present method, yet its utilization rate approaches only about fifty percent for large N. The symmetric algorithm extracts more from the transformed dataset family because the weight construction uses the full set as a coupled system. That is not a cosmetic computational detail. In absolute metrology, the value of an extra measurement often depends less on simply having more data than on whether the reconstruction actually knows how to use the symmetry those data carry. The present method does.  The experimental repeatability and the estimated sensitivity to rotation error keep the claims properly anchored. Rotational error below 0.10° left the rms behavior largely independent of N except at N = 2, and the actual experiment kept the angle error below 0.05°, placing its influence around 0.1 nm rms. That scale matters because it shows the algorithm is not floating in abstraction; it operates inside a realistic interferometric tolerance budget. To sum up, the study gives full-field absolute flat testing a more explicit symmetry logic, a sharper account of where residual structure enters, and a practical route to higher spatial fidelity with only four measurements. Beyond optical flat calibration, the method may also be relevant to semiconductor metrology, where nanometer-level full-aperture form measurement is increasingly important for wafers, wafer chucks, and precision reference surfaces.

Fig. 1. (a) Conceptional procedure for calculating the absolute surface shape using the symmetric algorithm. (b) Experimental setup for the wavelength-tuning Fizeau interferometer. (c) Measured absolute shape of flat C using the symmetric algorithm (N = 16), and the differences between the four algorithms (N = 2, 4, 8, and 16) and the classical three-flat test.

Gate errors accumulate long before a quantum processor reaches the scale needed for factoring, chemistry, networked entanglement, or error-corrected computation, and the mismatch is not subtle: present devices can already execute coherent operations and system demonstrations, yet the hardware stack still falls far short of the sustained, low-overhead performance demanded by those ambitions. That tension defines the scientific territory addressed here. In a recent review paper published in Science Journal, Professor David Awschalom and Dr. Hannes Bernien from University of Chicago together with Professor Ronald Hanson from Delft University of Technology and Professor William Oliver from Massachusetts Institute of Technology and Professor Jelena Vučković at Stanford University, developed a cross-platform hardware framework for quantum information processing that compares superconducting qubits, quantum dots, spin defects, trapped ions, neutral atoms, and photonic approaches through the shared demands of scalability. They identify four recurring hardware pressures—materials and fabrication, wiring, calibration and control, and size and power—and treat modular architecture as the organizing response. They also define quantum interconnects, heterogeneous integration, and photonic interfaces as the technical bridge from high-performing modules to large quantum systems. A central difficulty comes from the fact that quantum technologies have already left the stage of laboratory curiosities, yet the metrics that would let the community judge maturity in a rigorous and widely accepted way are still being formed. The authors make this point clearly when they discuss technology readiness across superconducting circuits, trapped ions, neutral atoms, photonics, quantum dots, and spin defects. Relative maturity can be observed, but high readiness in an emerging system does not mean that the end-state machine has arrived. That distinction matters because quantum hardware is now being pushed in operational environments, cloud platforms, sensing systems, and early network demonstrations, even as the raw performance needed for large-scale utility still demands a very different degree of control, reproducibility, and systems integration.

The review also argues that the unresolved character of this problem is not simply a matter of adding more qubits. Classical computing advanced through repeated shifts in base technology, fabrication practice, and systems-level co-design; the authors bring that history forward not as decoration but as a design lesson. Once device concepts were established, progress in electronics accelerated when system requirements began to guide materials and process development from the top down. Quantum hardware, in their reading, has reached the point where that same logic becomes productive. Bottom-up discovery remains valuable, yet scaling depends on choosing targets that are defined by the requirements of whole machines. That is the motivation running through the review. The question is no longer which isolated qubit can be made to function in a clean experiment. The harder and more consequential question concerns how one turns early platform success into extensible hardware: hardware that can be manufactured reproducibly, wired sensibly, calibrated in large numbers, powered within a realistic footprint, and linked into modules without losing the quantum character that gave the system value in the first place.

In their paper, the authors organized the evidence platform by platform, and that choice gives the review its practical force because it lets hardware performance be read together with the physical logic of each architecture. Superconducting qubits and lithographically defined quantum dots are treated as artificial atoms built by electrical design and operated at dilution-refrigerator temperatures. For superconducting processors, they describe arrays exceeding one hundred qubits, single-qubit fidelities in the 99.95 to 99.99% range, two-qubit fidelities around 99.5 to 99.9%, gate times of roughly 10 to 40 ns, and readout above 99% in 100 to 200 ns. Those numbers matter in context because speed compresses the error-correction cycle to about a microsecond, and that in turn makes large encoded demonstrations a meaningful systems test rather than a mere component benchmark. Their discussion of code distances three, five, and seven is especially telling: adding physical qubits reduced logical error rates stepwise, so scale itself began to function as an error-suppressing resource.

For semiconductor spin qubits, the review highlights a different design logic. Their lithographic compactness is attractive because it aligns naturally with industrial fabrication, and gate rates can approach the gigahertz regime. That same compactness also places pressure on wiring selectivity and cross-talk management, making control architecture part of the qubit problem rather than a peripheral engineering detail. The authors note single- and two-qubit operations at 99% fidelity, replication in a 300-mm foundry environment, three-nuclear-spin control above 99%, and four-qubit universal control in Ge/SiGe devices. They also point to valley splitting and operating temperature as physically meaningful materials parameters, since mixing with excited valley states directly feeds leakage and control error. A report of 99.8% Clifford fidelity at 1 K is framed not as a side result but as a sign that temperature itself can be reworked into the device strategy. Their treatment of spin defects, trapped ions, and neutral atoms shows how coherence, optical access, and modular networking are starting to converge. In solid-state spin systems, the researchers describe photonic interfaces, second-long electronic coherence, minute-long nuclear coherence, few-node networks, telecom-band frequency conversion, and entanglement through deployed fiber. In atomic systems, they emphasize the value of homogeneity: identical atoms simplify scaling because reproducibility is built into the physical object itself. Trapped ions exploit shared motional modes for high-fidelity gates, whereas neutral-atom tweezers convert optical field-of-view into a scaling variable and Rydberg excitation into an interaction mechanism over micrometer separations. That contrast is scientifically useful because it shows that “more qubits” is not a single route; each platform grows by leaning into its own physical resource.

Photonics threads through nearly every part of the review. The authors discuss all-photonic computing, cavity-QED cluster-state generation, and high-efficiency single-photon sources, then widen the frame to argue that photons are the preferred carriers for interconnects and multiplexed control across almost all hardware classes. This is not presented as a generic statement about light. It arises from the repeated experimental fact that long-range quantum linkage, modularity, trapping, readout, and chip-to-chip communication all begin to depend on optical functionality once systems move beyond monolithic demonstrations.

The scientific weight of the review comes from the way it changes the question. Rather than asking which platform is winning in a narrow sense, the authors ask what sort of hardware logic can carry quantum information technologies from impressive prototypes to extensible machines. That reframing matters because the review identifies a common structure across very different qubit modalities. Whether the qubit is a Josephson circuit, an electron in a gate-defined dot, an ion in vacuum, a neutral atom in a tweezer, a color center in a crystal, or a photonic cluster-state resource, the path forward begins to converge around a shared set of system demands: manufacturing discipline, scalable routing of signals, acceptable power and footprint, and calibration methods that remain tractable when qubit counts stop being small enough for manual tuning. That shift in emphasis has methodological value. It means quantum hardware can no longer be judged only by its best isolated fidelity number or by a striking demonstration in a carefully prepared setting. The review presses the community toward a more integrated standard in which device performance, architectural choices, and manufacturability must be read together. The authors’ appeal to the history of classical electronics sharpens this point. Progress in computing did not emerge from one magical device class maturing in isolation; it came from repeated coordination across materials, processes, circuits, and system requirements. By importing that lesson into quantum hardware, the review makes co-design feel less like managerial language and more like a scientific necessity.

A second implication concerns modularity. The review treats modular architecture as the route to the large performance jump that monolithic scaling alone is unlikely to deliver. This is a serious conceptual move. Once a machine is understood as an assembly of smaller units joined by quantum links, problems that looked overwhelming at whole-system scale become local design problems repeated many times. Wiring density, cooling burden, calibration load, and control complexity can then be capped at the module level. The price of that move is that interconnect quality becomes central. Fidelity, speed, and availability of the link start to shape the computer as much as the qubit itself. That is why the discussion of quantum memories, heralding, transduction, and hybrid interfaces carries so much weight. The authors are effectively arguing that future hardware may be defined as much by how subsystems talk to one another as by how any single subsystem computes. The review also gives photonics a broader scientific role than simple signal delivery. Photonic materials, couplers, waveguides, electro-optic elements, and single-photon sources become enabling pieces for modular quantum systems and, at the same time, a meeting point for quantum hardware and advanced classical interconnect technology. That convergence may shape future design culture in the area. Quantum hardware is starting to look less like a collection of isolated platform races and more like a heterogeneous engineering stack in which communication, memory, processing, sensing, and control can be optimized in distinct subsystems and then joined through carefully designed interfaces.

Random electromagnetic phenomena are not peripheral artifacts of physical systems; they are intrinsic to a wide range of natural and engineered environments, from cosmic background radiation to thermal noise and wireless communication channels. Despite their ubiquity, the manipulation of randomness has historically been treated as a circuit-level problem, addressed through electronic noise generation, filtering, or stochastic signal processing. In contrast, the electromagnetic wave domain itself—where randomness is carried, scattered, and transformed—has remained largely inaccessible to direct, programmable control. This disconnect has limited the ability to shape stochastic electromagnetic processes at their physical origin rather than at their electrical endpoint. Metasurfaces have emerged over the past decade as a powerful platform for electromagnetic wave manipulation, enabling compact, planar control over wavefronts, polarization, and spectral content. Digital and programmable metasurfaces, in particular, have introduced a level of flexibility that allows electromagnetic responses to be dynamically reconfigured in time. Most existing work, however, has focused on deterministic behavior: a known coding sequence produces a predictable field distribution. While this paradigm has enabled remarkable advances in beam steering, frequency conversion, and nonreciprocal wave control, it implicitly assumes that the desired electromagnetic outcome is ordered and repeatable.

However, many practical systems rely not on determinism but on controlled randomness. Communication channels are inherently stochastic, radar environments are clutter-dominated, and secure information transfer increasingly depends on unpredictable signal structures. Conventional random metasurfaces have demonstrated static random scattering patterns, but these structures lack temporal dynamics and cannot reproduce the statistical signatures of real stochastic processes. Crucially, they offer no systematic way to tune probability distributions, temporal correlations, or spectral bandwidth in a principled manner.  To this end new research paper published in Advanced Materials Technologies and conducted by Dr. Jia Cheng Li and Professor Tie Jun Cui from the State Key Laboratory of Millimeter Waves at Southeast University, the researchers developed a dynamically random coding metasurface that enables direct generation and control of stochastic electromagnetic signals in the wave domain. They introduced a probabilistic analytical framework linking random coding rules to field statistics, temporal correlation, and spectral characteristics. New mutation, inheritance, and shift coding models were experimentally validated, demonstrating tunable stochastic behavior without sacrificing randomness. This approach establishes metasurfaces as programmable platforms for engineered electromagnetic randomness rather than purely deterministic devices.

The research team implemented a dynamically random coding metasurface composed of a two-dimensional array of independently addressable meta-atoms and each element was designed as a 1-bit programmable unit capable of switching between two reflection phase states separated by π. This binary architecture, while intentionally simple, allowed large-scale collective behavior to emerge from many locally random events. A field-programmable gate array controlled the temporal evolution of the coding states, enabling rapid and repeated updates across the entire metasurface. Rather than imposing fixed patterns, the coding states were generated according to stochastic rules. In the simplest case, each meta-atom was assigned either state with equal probability at every time step, producing a uniform random code distribution. Measurements of the scattered electromagnetic field revealed that the resulting in-phase and quadrature components followed normal distributions, while the field amplitude obeyed Rayleigh statistics. These outcomes were not incidental; they reflected the central-limit behavior of many independent random contributions interfering in the far field. Importantly, these statistical properties were invariant with respect to observation angle, demonstrating that randomness was not confined to specific scattering directions.

The authors also monitored the scattered field over time, they quantified time autocorrelation and power spectral density, linking decorrelation speed directly to spectral bandwidth. Uniform random coding produced nearly instantaneous decorrelation and a flat spectral profile reminiscent of white noise. This established a baseline against which more structured stochastic behaviors could be engineered. They also explored three additional strategies that introduce structure into the temporal evolution of the metasurface while deliberately preserving randomness at the signal level. The first of these, mutation-based coding, allows each meta-atom to retain its current state most of the time, punctuated by occasional, probabilistic flips. This seemingly minor modification has a clear physical consequence: the scattered field decorrelates gradually rather than abruptly, and its spectral content shifts toward lower frequencies. In effect, the electromagnetic response begins to “remember” its recent past, albeit imperfectly and transiently.

Inheritance-based coding pushes this idea further. Instead of allowing every meta-atom to evolve independently, a subset of the previous spatial pattern is explicitly carried forward at each update, while the remaining elements are randomized anew. The result is a longer-lived temporal memory embedded directly in the wave domain. Experimentally, this means as slower decorrelation and a noticeably narrower spectral bandwidth, consistent with the increased persistence of spatial features across time.

Furthermore, the shift-based strategy introduces a different form of temporal linkage. Here, entire portions of the coding pattern are spatially translated before being refreshed. This operation produces a near-linear decay in temporal correlation, reflecting the deterministic motion imposed on an otherwise random field. Importantly, despite their distinct mechanisms, all three strategies produced experimental results that closely followed theoretical expectations, which reinforced the idea that stochastic electromagnetic behavior can be shaped using simple and physically transparent rules. One important observation is that none of these strategies reduced the intrinsic randomness of the signal. High entropy was consistently maintained, which indicates that temporal structure does not inherently compromise stochastic richness. When coding operations were combined, the metasurface showed even finer control over spectral shaping, which highlights the modular and composable nature of the framework.

In a nutshell, the reported study by Dr. Jia-Cheng Li and Professor Tie-Jun Cui demonstrates that randomness need not be treated as an uncontrollable byproduct of electromagnetic systems. Instead, it can be engineered deliberately, with mathematical rigor and experimental fidelity. The implications extend well beyond conceptual interest. In wireless communication, such controlled randomness offers a physically grounded route to realistic channel emulation. In cryptography, dynamically random electromagnetic fields provide a compelling basis for secure, hardware-level key generation. Radar and sensing systems, meanwhile, stand to benefit from programmable clutter and noise environments that more faithfully resemble real operational conditions.

Photovoltaic output departs from the weather-model expectation when sub-grid atmospheric processes and initial-condition errors distort the local irradiance history that actually governs panel response over the next few hours.   Grid operation needs forecasts that are short enough to reflect rapid meteorological change and stable enough to support dispatch, yet the data streams used for prediction do not carry the same kind of information. Numerical weather prediction supplies forward-looking meteorological sequences. Station instruments, on the other hand, record the local radiative and meteorological state directly, along with the plant’s power output, but they do not provide the future observations needed for actual forecasting. The central issue is not data quantity. It is that future model fields and past station measurements describe the same plant state on different time bases and from different observational standpoints. In a recent research paper published in International Journal of Electrical Power & Energy Systems, Professor Yuanbing Wang’s team from the Nanjing University of Information Science and Technology, developed a two-stage ultra-short-term photovoltaic forecasting framework that couples a CNN-BiLSTM short-term predictor with a CNN-Transformer revision model using cross-attention. The paper’s first author is student Yuchen Dai, with Professor Wang as the corresponding author. The research team also includes Professor Yaodeng Chen and students Junwen Wu and Jie Chao. Many plants already collect these station observations routinely, so the forecasting task can be recast around using them to revise a near-future sequence generated from weather-model input and in that setup, historical local measurements do not remain passive training data; they directly participate in correcting the short-term forecast.

They first used future meteorological model data in 24-hour sliding windows of 96 points to train a CNN-BiLSTM single-step predictor, producing a short-term output sequence for the coming day. They then extracted the first 4 hours of that forecast and aligned it with two station-side histories from the preceding 4 hours: measured meteorological observations and measured photovoltaic output.  A CNN-BiLSTM pathway suits the first stage because the convolutional layers capture local sequence structure and the bidirectional recurrent layer preserves dependence across the 24-hour input. The Transformer serves a different role in the second stage, where cross-attention lets the model treat the short-term forecast as a sequence to be revised using recent station history as the conditioning context. In the cross-attention block, the short-term prediction supplies the value vector, and the station-side histories supply query and key, so revision is driven by how the local past reweights the model-based future.

The source reports actual 15-minute data from four Jiangsu plants spanning April to December 2023, paired with ECMWF forecast data on the same interval. The team cleaned plant output with support vector regression using surface solar radiation as a reference variable and removed points above an empirically chosen error threshold of 80 W/(m²·sr). They also introduced a time-ratio feature marking each observation’s position within the daily cycle. That step matters because photovoltaic production follows not only weather variation but also the daily light cycle. Encoding time explicitly prevented anomalous nighttime generation in prediction and sharpened the periodic form of the output sequence. The sliding-window construction then preserved local temporal continuity for both the 24-hour and 4-hour problems, letting the two models operate on sequences built from the plant’s actual operating rhythm rather than on isolated samples.

The study contrasts five settings: a station-only 24-hour CNN-BiLSTM, a model-only 24-hour CNN-BiLSTM, a weighted fusion of those two outputs, a 4-hour CNN-BiLSTM built from matched station-side and model-side data, and the full hybrid 4-hour method. Averaged across the four stations, the proposed model reached an MAE of 3.41 and an RMSE of 6.21, compared with 4.09 and 8.57 for the model-only 24-hour baseline. That corresponds to reductions of about 16.7% in MAE and 27.4% in RMSE. Relative to output weighted fusion, the reductions were about 16.8% and 27.9%; relative to the 4-hour CNN-BiLSTM, they were about 7.7% and 20.1%. Station-level results followed the same pattern: MAE and RMSE fell by about 16.22% and 27.42% at Station A, 23.29% and 33.20% at Station B, 20.38% and 29.34% at Station C, and 6.83% and 14.59% at Station D. The paper further notes that the hybrid forecasts tracked peaks more closely under stable weather, followed hourly variation with greater detail, and improved the timing of sunrise and sunset transitions. Peak shape and day-edge timing directly affect any calculation that integrates power over time, so better local alignment changes the practical content of the forecast itself. After the researchers computed Pearson correlations among meteorological variables and power output, they built four groups, starting with the two most correlated radiation variables and then expanding with clouds, temperature, visibility, wind, and additional radiation terms. Group 3, which combined ssrd, ssr, tsr, tsrc, ssrc, fdir, cdir, mcc, tcc, t2m, vis, ws, and wd, produced the best ultra-short-term metrics, with MAE 1.3492 and RMSE 3.0049. The new work interprets this result by showing that strongly positive radiation variables contributed most effectively during ultra-short-term revision, where they shaped the modal changes of the forecast sequence. Fine-grained radiative detail becomes most useful after the forecast has already been formed, when the model needs to adjust short-horizon dynamics rather than build the entire day from scratch.

The new research paper changes the forecasting logic from direct prediction to forecast revision driven by recent plant-specific evidence. In many photovoltaic workflows, meteorological forecast data serve as the main predictive substrate, while station observations remain historical inputs. Here, the short-term forecast becomes an intermediate physical statement rather than a terminal answer. Ultra-short-term prediction then becomes a correction problem in which recent plant behavior reshapes the near-future sequence generated from weather-model input. The CNN-BiLSTM stage compresses and propagates temporal structure from future meteorological model input across a 24-hour horizon. Its cross-attention mechanism lets local station histories interrogate the already-formed forecast. It expresses two distinct sources of information in photovoltaic prediction: the broad meteorological trajectory supplied by the forecast model and the near-field operating state recorded at the plant. Keeping those roles distinct makes the revision interpretable in functional terms. The plant-side sequence is not asked to predict the future by itself, which would conflict with the absence of future observations. Instead, it modifies a forecast that already exists, and that is precisely the regime where local measurements carry the most value.

Radiation variables with strong positive correlation to power output, joined with cloud, temperature, visibility, and wind information, proved especially effective in the revision stage. The paper’s own interpretation is that these variables enrich the dynamic response of the ultra-short-term forecast. That matters because photovoltaic output is shaped not only by bulk irradiance magnitude but also by how radiative and meteorological factors vary over short intervals. This framing assigns short-scale radiative variation a distinct role during forecast revision rather than during initial sequence construction. The study, then, contributes a way of thinking about feature utility: some variables may be most informative not at the stage of initial day-ahead sequence construction, but at the stage where short-horizon corrections are made to evolving local behavior. The study also frames the method as portable. It states that deployment in a new region requires three inputs: historical weather observations, historical photovoltaic generation data, and numerical weather prediction fields obtainable from geographic coordinates, with adaptation handled through parameter fine-tuning rather than architectural redesign.   A method that depends on widely available plant records and standard forecast products fits naturally into operating environments where bespoke instrumentation is not the main route to better forecasting. What the study contributes, then, is a hybrid revision framework for ultra-short-term photovoltaic prediction in which recent station evidence and forecast-model sequences are not competing alternatives. They become coordinated layers of information, each used at the stage where it carries the greatest explanatory force for the next few hours of plant output.

Scientists have made chemical systems smaller and smaller (down to the micro-scale) the level of control and precision we can now achieve in these microfluidic systems is far higher than what people thought was possible in the past. Once reactions are confined to channels only a few hundred micrometers across, even small disturbances in flow begin to matter in ways they simply do not in larger systems. A slight drift in velocity might reshape a concentration gradient, alter how long a reagent lingers in a zone of interest, or nudge a reaction along a different path altogether. For this reason, flow sensing has gradually shifted from a convenience to a necessity within microfluidic design but the tools we rely on have not evolved at the same pace. Most commercial sensors still sit outside the device, bolted onto tubing rather than integrated into the chip itself. This arrangement limits where measurements can be taken and, in many setups, consumes space that researchers would prefer to use for additional microfluidic functions.

Another complication arises the moment these platforms are adapted for chemical synthesis or analytical work. The palette of usable materials narrows quickly. Many MEMS-friendly substrates—silicon, PDMS, other metals and polymers—behave nicely from a fabrication standpoint but are far less cooperative in chemically demanding environments. They may leach, swell, or participate in redox reactions after extended contact with solvents or reactive intermediates. Glass avoids most of these issues. Its chemical stability and optical clarity make it a natural choice for high-precision microfluidics. Even so, embedding MEMS sensing elements within a glass architecture is not trivial. The conductive components must remain electrically insulated from the flowing liquid, and the fabrication tolerances needed to maintain reliable flow control leave little room for error. To this end, new research paper published in Journal of Micromechanics and Microengineering and conducted by Mr. Shao-Yang Hung, Mr. Zhong-Wei Lin, Professor Hiroyuki Fujita, Professor Sheng-Shian Li, Professor Chihchen Chen, Professor Takehiko Kitamori and Professor Kyojiro Morikawa from the National Tsing Hua University, the researchers designed and fabricated a noncontact thermal MEMS flow sensor embedded directly into a glass microfluidic chip, ensuring full chemical compatibility by isolating all metallic elements behind a thin glass wall. Their system integrates a nickel heater and paired upstream–downstream sensing wires to create a calorimetric measurement scheme capable of detecting flows as low as 0–8 µl min⁻¹. They developed and validated two complementary models—one based on multiphysics thermal simulations and one based on experimental demodulated electrical signals—that jointly describe heat transport and sensitivity limits in glass-insulated microchannels.

The researchers patterned a nickel heater and two nickel sensing wires on a thin glass substrate positioned above a machined channel. The channel itself measured 550 µm in width and 100 µm in depth, dimensions chosen to maintain manageable thermal diffusion while offering compatibility with standard microfluidic operations. A second glass layer, only 170 µm thick, separated the fluid from the metal elements, forming the mechanical and electrical insulation required for chemical work. Fabrication proceeded through photolithography and lift-off, followed by the bonding of glass layers, careful CNC machining of channels and ports, and wire bonding to a flexible cable for electrical interfacing. Afterward, the authors used multiphysics modeling to anticipate how heat would propagate through the multilayer structure. They adjusted heater dimensions, sensing wire placement, and boundary conditions until the predicted thermal field could reliably distinguish upstream from downstream temperature changes. These simulations included contributions from natural convection, conduction through the glass wall, and the effects of flow velocities spanning the microliter-per-minute range. They found the selected design maintained a stable gradient at a 20 mA heating current, and allowed detectable shifts in resistance through the temperature-dependent response of nickel.

Experimentally, the team mounted the glass chip in a custom holder connected to a syringe pump. An AC-modulated signal was applied to the sensing wires and demodulated through a lock-in amplifier to extract temperature-dependent resistance changes with minimal noise. They found at low flow rates, the downstream wire consistently registered an elevated temperature while the upstream wire cooled slightly, producing a differential signal that increased monotonically with flow. The linear regime extended from 0 to 8 µl min⁻¹, a region where conventional flow sensors typically struggle. Beyond this range, the downstream signal began to plateau, a behavior the authors traced to heat loss into the channel walls and imperfect downstream heat transport. The authors also showed negligible hysteresis when flow rates were ramped up or down, suggesting that the device’s thermal response remained stable over time and did not retain a memory of previous flow conditions. Moreover, the sensitivity values were modest when calculated in terms of volumetric flow rate—an expected outcome given the small cross-section—but comparable to or better than those of earlier MEMS devices when assessed in terms of flow velocity. Moreover, the team noted that increasing heater power could enhance sensitivity, though at the cost of elevated fluid temperatures, an issue of particular importance in chemical or biological assays.

In conclusion, the new work of Professor Kyojiro Morikawa developed new models that establish a framework for future high-precision, chemically robust flow sensors suitable for advanced lab-on-chip applications. The authors resolved the chemical incompatibility of many MEMS materials by embedding the MEMS components directly into an all-glass architecture. Their strategy of isolating metallic structures behind a thin glass barrier preserves the advantages of calorimetric sensing while eliminating concerns about contamination or electrochemical interference. This is particularly relevant for applications where even trace metal ions or unintended redox reactions could distort analytical measurements or compromise product purity. The study also advances our understanding of thermal flow sensing at small scales. The linear response observed within the 0–8 µl min⁻¹ range demonstrates that calorimetric techniques remain effective when carefully adapted to the thermal properties of glass. Their findings highlight the delicate balance between heat conduction through the substrate and heat convection within the channel and that too much insulation weakens the signal; too little risks overheating the analyte. Additionally, the authors’ decision to work with a 170 µm glass separation illustrates an attempt to reconcile these competing pressures, though their own discussion points toward future opportunities in ultra-thin glass technologies. Sheets only a few micrometers thick could dramatically sharpen thermal coupling while maintaining chemical inertness, offering a pathway toward even more sensitive, lower-power sensors.

Moreover, instruments used for synthesis, catalysis screening, biological assays, or reaction monitoring increasingly rely on internal feedback to regulate conditions in real time. External flow meters are often too sluggish or too coarse for such tasks. The demonstrated ability to integrate multiple thermal sensors along a single channel—fifteen units in this device—suggests that spatially resolved flow monitoring could become routine. This would allow researchers to observe local flow disturbances, diagnose channel blockages, or quantify reaction-induced viscosity changes without modifying the chemical environment. Another significance is the reproducibility and the absence of hysteresis ensures that the device does not require recalibration between measurements which simplifies its deployment in automated systems. Its all-glass construction also aligns with the needs of optical detection methods frequently used in microfluidics, offering compatibility with fluorescence, absorbance, and imaging techniques. In sum, the new work provides a blueprint for the next generation of chemically compatible, chip-integrated flow sensors opens avenues for integrating multi-sensor arrays into increasingly complex microfluidic platforms.

The bulk of hydrogen still comes from natural gas, which obviously limits its environmental value. As more governments and industries talk seriously about reducing emissions, the spotlight has shifted toward ways of generating hydrogen directly from renewable electricity. Among those, pairing photovoltaics with electrolysis is the option that feels the most grounded in existing technology. You take sunlight, convert it to electricity, and split water. On paper, it’s almost disarmingly simple. In practice, especially when scaled beyond a modest test setup, things start to behave less cleanly. One issue is that PV arrays never sit still electrically. Their operating point wanders as sunlight changes or as the panels heat up. Engineers normally handle this wandering with power converters running maximum-power-point tracking (MPPT), which continuously nudges the array toward its most productive point. That works beautifully for most solar applications. Electrolyzers, however, care about something slightly different. Their hydrogen output depends on current flow, not electrical power. So an algorithm optimized for power isn’t necessarily the one that delivers the most hydrogen. This mismatch becomes even more glaring when you look at the current requirements of modern electrolyzer stacks—currents so large that existing DC/DC converters simply aren’t built to handle them. Large installations often fall back on multi-stage AC coupling, and with that comes extra hardware, extra losses, and extra expense. Because of those complications, direct coupling (wiring the PV array straight into the electrolyzer) has drawn renewed attention. Removing converters avoids their efficiency losses and their current limitations, which is appealing. But there’s a catch: without any form of regulation, the PV voltage drifts wherever the sunlight pushes it. Electrolyzer stacks aren’t indifferent to voltage; running too far outside their recommended window shortens their lifetime. Earlier attempts at direct coupling did occasionally land near good operating points, but only in a passive way. They weren’t designed to favor hydrogen yield, and as a result, they often left quite a bit of performance on the table.

To this end, new research paper published in International Journal of Hydrogen Energy and led by PhD intern Kelvin Tan, and Professor Meng Tao from the School of Electrical, Computer, and Energy Engineering at Arizona State University, the researchers developed a load-matching photovoltaic–electrolyzer system that operates without any conventional power converter, relying instead on relay-based switching of electrolyzer stacks to adjust load impedance. Within this architecture they designed a maximum current point tracking (MCPT) algorithm that prioritizes hydrogen-producing current while maintaining voltages inside the electrolyzer’s preferred operating range. The algorithm consistently delivers more charge than classical MPPT and dramatically improves voltage compliance.

The authors approached the problem through a blended simulation–experimental framework in which the electrical behavior of a four-stack proton-exchange-membrane (PEM) electrolyzer system was coupled directly to a modeled 400-kW PV array. Instead of placing any power converter between the two subsystems, each stack could be switched on or off via relays, altering the overall load impedance seen by the PV source. By sweeping through combinations of connected stacks, the system generated families of operating points that depend on time-varying irradiance. These operating points, shown across power, current, and voltage domains, provided the foundation for identifying which combinations delivered either maximum power or maximum current at any given moment. The experiments unfolded in two phases. The first was a detailed simulation driven by irradiance data from Tucson, Arizona. To make the problem tractable while preserving the dynamics of a full day, the irradiance profile was compressed in time. Under this input, the MCPT algorithm was tested against the best available MPPT method developed previously by the group. Both algorithms respond by connecting or disconnecting stacks as sunlight strengthens or weakens. However, the MCPT algorithm applied additional logic: every prospective switch had to be evaluated for whether it increased current while still preserving voltages within a defined “safe” window. If a switch produced higher current but pushed the stack out of the acceptable voltage range, the algorithm reverted to the previous configuration.

The new approach created notable behavioral differences between the two algorithms where in early morning, when irradiance is low, MCPT tends to connect additional stacks earlier than MPPT, accepting a modest power penalty in exchange for higher current. In the afternoon, a symmetric pattern appears, with MCPT retaining a higher number of connected stacks for longer. Across the full day, these shifts accumulated into a clear trend: MCPT delivered roughly 2.5% more total charge than MPPT. The authors conducted voltage statistics and observed with its voltage-aware switching logic, MCPT kept the system within the electrolyzer’s recommended voltage range around 72% of the time which is substantially better than MPPT. Additionally, when they introduced an additional constraint that allow the system to begin the day with zero connected stacks further improved voltage compliance to roughly 94% was observed, albeit with a small reduction in total charge. In the second phase, a physical proof-of-concept system was assembled using a 290-Wp module and resistor–Zener assemblies tuned to mimic PEM stack behavior. Over both sunny and cloudy days, the MCPT algorithm demonstrated the same qualitative features observed in simulation: current was consistently prioritized, voltage remained stable within the optimal window, and the number of connected stacks adjusted responsively to rapid irradiance fluctuations. Even though real-world temperature variations created small impedance mismatches, the general consistency between simulation and experiment reinforced the credibility of the algorithm.

In conclusion, Kelvin Tan and Meng Tao successfully developed new MCPT algorithm that present a low-cost, converter-free route for scalable solar-driven hydrogen production. Indeed, the new study makes a strong case for rethinking how photovoltaic systems intended for hydrogen production should be controlled and engineered. Traditional PV‐electrolyzer integration is shaped by the logic of power electronics: regulate voltage, maximize power, and let the electrolyzer accept whatever current results. But as the authors show, this mindset becomes counterproductive once hydrogen yield becomes the metric of interest. A system that extracts slightly less electrical energy from the PV array can nonetheless generate more hydrogen if it maintains higher current over larger portions of the day. The MCPT algorithm places this insight into a practical form by recasting control decisions as current-maximizing moves that still respect voltage boundaries. It is noteworthy to mention the benefits of MCPT arise without adding any new hardware. Direct coupling retains its appeal because every removed converter translates into lower cost and higher reliability—factors that matter enormously when imagining hydrogen plants made of dozens or hundreds of megawatt-scale stacks. The algorithm also addresses a longstanding vulnerability in direct-coupled systems: voltage drift. Previous demonstrations often accepted wide voltage swings as an unavoidable feature of simplicity. By embedding voltage regulation directly into the switching logic, Tan and Tao create a version of direct coupling that behaves in a more disciplined, industrially acceptable manner. The improvement in voltage compliance—from less than half the day under MPPT to nearly the entire day under the revised MCPT which is substantial enough to influence decisions surrounding stack durability, replacement frequency, and overall lifecycle cost.

A broader implication lies in how PV arrays and electrolyzers might be co-designed. Instead of viewing power electronics as the universal interface, this work hints at a future in which electrolyzer geometry, number of stacks, and PV string configuration can be chosen jointly, with the load-matching algorithm acting as the glue. The approach scales naturally because it depends only on how many stacks the relays can address. This is appealing for gigawatt-level projects where current levels will far exceed what today’s converters can handle. In such environments, designing systems to deliver high current directly from the PV array could sidestep engineering bottlenecks that otherwise limit deployment speed. If integrated with new stack chemistries or high-current PEM designs, MCPT-like strategies could fundamentally change how future hydrogen farms are architected.

Satellite electrical power systems are being asked to do more than they were ever originally designed for. Missions are longer, payloads are heavier, and subsystems that once operated independently are now tightly interwoven. Under these conditions, electrical power regulation is no longer a background engineering concern; it increasingly determines whether a satellite performs as intended over its full operational lifetime. The power conditioning unit (PCU) mediates the exchange of energy between solar arrays, batteries, and onboard loads, and it does so continuously, under changing illumination, temperature, and demand and capturing the behavior in a model that remains faithful to on-orbit reality is still surprisingly difficult. Most prior modeling efforts fall into one of three categories. Physically derived circuit models emphasize interpretability and control logic, but they often require simplifications that smooth out precisely the dynamics that matter most during mode transitions or disturbances. Data-driven approaches move in the opposite direction, and they offer flexibility and fast fitting at the expense of physical transparency. These models can perform well in narrowly defined regimes, however, they tend to degrade when operating modes shift, and they rarely provide firm guarantees on stability. Hybrid methods attempt to reconcile these shortcomings, but in practice they introduce layers of complexity that limit their usefulness outside offline analysis. As a consequence, many PCU models remain detached from real operational contexts and are poorly suited for digital twin implementations. Part of the difficulty lies in the PCU itself. Its operation spans multiple domains simultaneously. Control behavior changes between illumination and umbra, depends on interactions among several regulators, and unfolds across distinct frequency ranges. Time-domain models are effective at reproducing transients, but they often struggle to sustain fast and stable regulation when system conditions evolve. Frequency-domain approaches, by contrast, are well suited for stability analysis, however, they are frequently applied only to isolated subsystems or simplified operating modes. The result is a fragmented modeling landscape that fails to represent the PCU as a coherent whole. Moreover, another challenge is the loose connection between models and actual satellites and validation is commonly based on semi-physical experiments or functional benchmarks, which provide limited insight into how models behave once environmental variability and telemetry noise are introduced. Without continuous alignment to on-orbit data, even carefully constructed models tend to drift over time. Closing this gap between analytical fidelity and operational relevance remains one of the central challenges in PCU modeling. To this end, new research paper published in Aerospace Science and Technology and conducted by PhD candidate Meng Wang, Professor Guangquan Zhao, and Professor Xiyuan Peng from the Harbin Institute of Technology, the researchers developed a high-fidelity, frequency-domain model of a complete satellite power conditioning unit that operates in real-time synchronization with on-orbit telemetry. Their framework integrates shunt regulation, battery charging, and battery discharging into a unified control architecture optimized across mid- and high-frequency bands.

The research team adopted a physically grounded frequency-domain modeling strategy, in which the complete PCU is treated as a closed-loop, multi-module control system rather than a collection of loosely connected subsystems. The modeling process begins by establishing energy-flow relationships among the error amplifier, shunt regulator, battery discharge regulator, and battery charge regulator. These relationships define a unified control-loop backbone that governs power regulation across both illumination and umbra periods. For the illumination phase, the shunt regulator and main error amplifier are modeled as a coupled feedback system. The shunt hysteresis behavior, traditionally treated as nonlinear and difficult to analyze, is linearized through spectral approximation, enabling its representation as an equivalent proportional link. This simplification allows magnitude and phase characteristics to be explicitly optimized in the mid-frequency range, where regulation sensitivity is highest. Parameter selection is guided by crossover frequency and phase margin targets, yielding a stable system with rapid convergence and minimal ripple in bus voltage. During umbra periods, attention shifts to the battery discharge regulator, implemented through a boost-type DC–DC converter. Here, the authors employ state-space averaging to derive a small-signal frequency-domain model that captures the regulator’s non-minimum-phase behavior. Crucially, the effect of equivalent series resistance is retained rather than neglected. The authors demonstrated that this resistance plays a stabilizing role, compensating right-half-plane zeros and suppressing resonant peaks that would otherwise destabilize the control loop. Compared with models that omit this effect, the resulting system exhibits markedly improved phase and gain margins. They treated the battery charge regulator differently because it does not directly regulate the power bus, a simplified equivalent model is adopted to reduce computational burden without sacrificing accuracy in overall system behavior. The authors’ design choice reflects a pragmatic balance between fidelity and efficiency, particularly important for real-time simulation. The team validated using 24 hours of real telemetry data from a geostationary satellite, sampled at sub-second resolution. Model outputs—including bus voltage, battery voltage, charging and discharging currents, solar array current, and error amplifier voltage—are continuously compared against measured data. Across all metrics, the proposed model demonstrates substantially lower mean absolute, mean squared, and root mean squared errors than existing approaches. In several operating modes, simulated outputs are nearly indistinguishable from telemetry. Moreover, the authors conducted introduced mode transitions, reference offsets, and battery cell faults and found in each case, the system stabilizes rapidly, with settling times well below one second and overshoot confined within acceptable limits.

In conclusion, the research work of Harbin Institute of Technology scientists successfully developed a new model that can achieve both superior dynamic regulation and unprecedented agreement with real satellite data by coupling spectral parameter optimization with continuous telemetry feedback. This innovative approach effectively transforms the PCU model into a functional digital twin capable of supporting on-orbit operation and fault management. Additionally, the study challenges the prevailing reliance on time-domain simulations for PCU analysis and although such models are valuable for capturing transients, they are limited in delivering stable dynamic regulation across changing operating modes. In contrast, the frequency-domain approach adopted by the authors integrates transient and steady-state behavior within a single framework, which enable control over stability margins and dynamic response and this shift has important implications for how future satellite power systems may be designed and validated. Moreover, the ability to synchronize the model continuously with on-orbit telemetry represents a major advance and instead of serving as a static design artifact, the PCU model becomes an active participant in system operation, capable of tracking real conditions, and can predict near-term behavior, and support informed decision-making. This capability is valuable for fault diagnosis and health management, where early detection and rapid response can extend mission lifetimes. The results also highlight the importance of modeling completeness. By incorporating the interactions among all major PCU modules, the proposed framework avoids the blind spots that arise when individual regulators are treated in isolation. The dramatic reduction in modeling error across multiple output parameters highlights the value of this holistic perspective. We believe, the implications extend beyond geostationary satellites and sequential switching shunt regulators and although the present study focuses on a specific architecture, the underlying methodology—frequency-domain optimization coupled with telemetry-driven synchronization—can be generalized to other satellite platforms and power system designs. Moreover, the authors’ acknowledgment of remaining discrepancies points toward a natural next step: integrating data-driven correction layers, such as neural networks, to further refine physical models without undermining their interpretability. In a nutshell, the study by Harbin Institute of Technology scientists developed a highly accurate PCU model, that can guide how future aerospace systems might be modeled, monitored, and managed on orbit.

When photonic platforms are used to handle microwave signals, the system inherits wide optical bandwidths, minimal transmission loss, and a welcome immunity to electromagnetic noise. These benefits have become increasingly important as radar, wireless communication, and data-processing systems push toward higher frequencies and larger dynamic ranges. Recently, improvements in integrated photonic fabrication especially silicon photonics and a range of heterogeneous material stacks have broadened what designers can actually implement on chip. Functions such as carefully shaped microwave filters, chip-level differentiators, and compact arbitrary-waveform generators are now realistic building blocks for full systems. However, despite the sophistication of these platforms, the way microwave photonic systems are designed remains surprisingly traditional. Most researchers approach a new function by reaching for well-known components—FIR filters for generality or IIR structures based on microring resonators when high-Q shaping is required. Both choices come with limitations that everyone in the field is aware of: FIR filters can approximate almost anything if one is willing to accept a long tap chain, and IIR filters stay compact but inevitably restrict the frequency response to a narrow family of shapes. When the two are hybridized, the design space widens a bit, though not nearly enough to explore all the architectures that are mathematically possible. As a result, many potentially superior configurations most likely sit untouched, simply because they do not resemble the circuit motifs that the field has grown comfortable with. The situation becomes more complicated as photonic chips incorporate larger routing networks, more elaborate cascades of resonators, and tunable elements that change the system’s operating point. Once these features are available, the number of feasible architectures expands extremely quickly—so quickly that intuition offers little guidance about where the “good” ones might lie. Attempting to sort through these possibilities manually becomes unrealistic, and even experienced designers cannot easily tell whether a proposed configuration is anywhere near optimal. In practice, this creates a widening gap between what integrated photonics enables and what conventional microwave-photonic design methods can systematically explore.

 To this account, new research paper published in Optics Express and conducted by Mr. Bo Li, Prof. Shaofu Xu, Ms. Ting Lyu, Mr. Ruicheng Qiu, Ms. Yixi Liu, and Prof. Weiwen Zou from the Department of Electronic Engineering at Shanghai Jiao Tong University, researchers developed an automated design framework that represents microwave photonic systems as complete topological graphs and searches this space systematically. They combined exhaustive topology generation with a hybrid genetic-algorithm and L-BFGS parameter search to identify architectures that match target responses with minimal system cost. The method consistently uncovers non-intuitive designs that outperform conventional FIR and IIR configurations. Across differentiators, waveform generators, Hilbert transformers, and pulse-compression systems, the framework delivers architectures that are simultaneously more efficient and more accurate than existing hand-designed counterparts.

The authors defined a structured way to represent any microwave photonic system as a graph. Each processing block—whether an FIR filter or one of the IIR variants—is treated as a node with tunable parameters. Connections between nodes form the edges, encoding whether units operate in series or in parallel. Once this abstraction is established, the team develops an exhaustive graph-generation routine that enumerates all feasible directed acyclic graphs constructed from a chosen number of basic units. The algorithm then converts each graph into an associated analytical expression using depth-first search, ensuring that the full diversity of transfer functions emerging from these combinations is preserved. Redundant expressions are removed to avoid unnecessary computation during later stages.  The researchers then turned to parameter optimization. Here, the task is to tune the coefficients, delays, and loop gains of the processing units so that the resulting system response matches a target function. Depending on the application, the target may be an ideal mathematical operator—such as a temporal differentiator—or an output waveform directly, as in the case of arbitrary waveform generation. To balance accuracy against hardware complexity, the loss function includes both a mean-square-error term and a cost term proportional to the number of tunable parameters inside the architecture. Minimizing this loss requires navigating a high-dimensional, non-convex space, so the team employs a hybrid scheme: a genetic algorithm performs a global search, and the L-BFGS method refines promising candidates to convergence.  They applied the workflow to several benchmark functionalities and found for the second-order temporal differentiator, the automated search not only rediscovers the known cascaded IIR configuration but also produces a non-intuitive alternative that achieves lower error with fewer parameters. A similar story unfolds with the Hilbert transformer. The framework identifies an architecture capable of matching the desired π/2 phase shift while reducing amplitude ripple relative to the conventional 12-tap FIR design. In both cases, the system’s output for Gaussian and sinusoidal test signals demonstrates that these automatically derived architectures behave exactly as required. Furthermore, the authors reported for sawtooth, triangular, and square waveform generation, the search discovers compact structures formed from multiple FIR filters arranged in mixed series–parallel layouts. These achieve lower mean-square errors than the standard single-filter approach and do so with reduced tap counts. Similar gains are observed in linear-frequency-modulated pulse compression, where a newly found architecture significantly improves peak side-lobe ratio while using less than half the parameter budget of the conventional 20-tap FIR implementation. Even in millimeter-wave signal generation, the algorithm selects a single bandstop IIR filter as the optimal spectral shaper, cutting system cost and also preserving side-mode suppression. The team also conducted a robustness study which further showed that these architectures remain resilient under typical fabrication errors, and retain performance advantages over their conventional counterparts.

In conclusion, the new work by Shanghai Jiao Tong University scientists demonstrated that microwave photonic architectures can be discovered automatically, the authors open the door to a fundamentally different design culture—one in which intuition serves as a starting point but no longer dictates the boundaries of what is possible. Their results show, often dramatically, that optimal solutions do not always resemble familiar structures. In some cases, the best architecture contains unexpected combinations of parallel and series paths; in others, it relies on a deliberately minimal set of components whose utility only becomes apparent once the loss landscape is mapped numerically. The new automated framework evaluates complexity explicitly through its cost term, rewarding architectures that achieve performance targets with fewer adjustable parameters. This naturally favors solutions that remain compact and physically implementable on integrated platforms and with the photonic chips continue to grow in scale and functionality, this cost-aware optimization will become critical, especially when designers must accommodate tight power budgets, limited chip area, or stringent thermal constraints.

Scalability is an important advantage of the new design and the graph-based representation provides a direct pathway toward more sophisticated systems, including multi-input multi-output configurations or nonlinear photonic processors. Although the study focused on linear, single-input single-output systems, the framework itself is not inherently limited to these cases. Adjustments to node types and loss definitions could, in principle, enable automatic discovery of architectures for optical neural networks, multi-channel sensing systems, or photonic compute blocks. The ability to explore these expanded design spaces algorithmically could accelerate progress across a spectrum of emerging technologies that are beginning to rely on photonic signal processing. Another advantage is the framework’s robustness to fabrication imperfections. The authors show that even when amplitude and phase errors of several percent are introduced, the automatically discovered architectures maintain performance advantages over conventional solutions which suggests that the framework produce mathematically optimal designs and also as important identifies physically resilient ones. In sum, the authors create a toolbox capable of uncovering architectures that combine efficiency, performance, and implementation. As microwave photonics continues expanding into areas such as high-capacity wireless communication, cognitive sensing, and photonic computing, the impact of such a methodology is likely to grow.

There is an increasing demand for compact, high-power, and highly linear RF transistors continues to escalate next-generation wireless communication and radar systems. Gallium nitride (GaN)-based devices, particularly high electron mobility transistors (HEMTs), have emerged as the cornerstone of this technological evolution. Their intrinsic properties—wide bandgap, high breakdown voltage, and superior electron transport—make them uniquely suited to endure the harsh electrical stresses imposed by high-frequency and high-power environments. Yet, even with these advantages, there remain fundamental obstacles that constrain the full exploitation of GaN in advanced RF circuitry. Chief among these is the need to enhance power density without compromising linearity or reliability, a balancing act that has proven difficult as devices are pushed closer to their material limits. One promising route to resolving this impasse lies in the use of multichannel architectures. Among these, the superlattice castellated field-effect transistor (SLCFET) stands out. Unlike traditional single-channel GaN HEMTs, the SLCFET incorporates vertically stacked two-dimensional electron gas (2DEG) channels formed via an AlGaN/GaN superlattice structure. This enables a dramatic increase in charge carrier availability without enlarging the device footprint. By combining this dense channel configuration with a three-dimensional fin geometry and wrap-around gate control, SLCFETs offer a compelling path toward ultra-compact, high-power RF transistors. However, the complexity of this structure introduces new variables—particularly fabrication-induced variation in fin dimensions—that can influence device behavior in subtle but significant ways.

New research paper published in Nature Electronics and led by Professor Martin Kuball from the University of Bristol and conducted by Dr. Akhil Kumar, Stefano Dalcanale, Michael Uren, James Pomeroy, Dr. Matthew  Smith alongside Justin Parke and Robert Howell from the Northrop Grumman Mission Systems in the United States, the researchers developed a multichannel GaN SLCFET that exhibits a latch-induced subthreshold slope below the 60 mV/decade Boltzmann limit. They demonstrated that localized impact ionization in the widest fins initiates reversible latching, leading to enhanced transconductance linearity without degrading device reliability. This architecture leverages naturally occurring fin-width variations to improve RF performance, offering a novel, scalable approach for high-power applications.

The researchers began by fabricating multifinger SLCFETs incorporating ten stacked AlGaN/GaN superlattice periods, each forming a distinct 2DEG channel. These channels were electrostatically modulated through conformally coated SiN dielectric and wrap-around gate metals on densely packed nanoscale fins. Through initial current–voltage (I–V) sweeps, they observed a peculiar two-phase subthreshold behavior—one segment with a gradual 80 mV/decade slope and another where the current surged abruptly with a slope well below the thermionic limit. Notably, transconductance (gm) plots revealed multiple shoulders, which hinted at nonuniform channel activation that could not be explained by standard GaN HEMT physics.

To trace the origin of these irregularities, the team utilized scanning electron microscopy (SEM) and found significant variation in fin widths across the device, a byproduct of fabrication processes. They then turned to 3D simulations using Silvaco Atlas to model how fin-width dispersion impacts device performance. Simulating fins with a ±20% width deviation around the mean, they reproduced the same subthreshold gm shoulders observed experimentally. This confirmed that wider fins, due to their more negative threshold voltages, entered conduction earlier, effectively initiating a cascade of channel activation at distinct gate voltages. However, the most intriguing discovery came when they performed electroluminescence (EL) microscopy during subthreshold I–V sweeps at high drain bias. As the gate voltage was stepped gradually from −12 V to −10.5 V, faint but highly localized EL spots emerged, aligned with abrupt increases in drain current. These hot spots—originating near the gate-drain edge—provided direct visual evidence of impact ionization occurring in individual fins. Crucially, as the sweep continued, additional EL spots appeared, suggesting that more fins were sequentially entering a latched, high-conduction state. Afteerward, the team conducted repeated bidirectional gate sweeps under high-field conditions. The result was both surprising and reassuring: the sharp switching behavior remained stable over 90 minutes with no signs of degradation. Gate current profiles further supported a recoverable process, showing a bell-shaped peak consistent with hole generation and subsequent release. Simulated band diagrams reinforced the hypothesis that holes temporarily accumulated at a barrier formed between GaN and residual native oxides at the fin-dielectric interface, altering the local threshold voltage. The authors compared transconductance before and after latching and demonstrated a broadened gm profile and a marked reduction in second-order nonlinearity (g″m). These outcomes strongly indicated that the latching effect, rather than undermining performance, actually enhanced the device’s RF linearity—a rare instance where a seemingly erratic behavior turned out to be a hidden asset.

In conclusion, Professor Martin Kuball and colleagues successfully revealed that under controlled conditions, it could produce ultrasteep subthreshold slopes and improve linearity in high-frequency applications. This challenges prevailing assumptions in device engineering, particularly for GaN, where the high bandgap has long been thought to preclude such behavior without structural degradation or reliability risks. Indeed, the discovery that a single wide fin—formed unintentionally during standard lithography—can initiate reversible latching through localized impact ionization represents a turning point. It underscores the importance of embracing rather than eliminating geometrical nonuniformities, provided their influence can be accurately understood and modeled. This paradigm shift could open new pathways for device tuning by harnessing naturally occurring variability, rather than expending resources to eliminate it entirely. From a practical standpoint, the study carries direct implications for next-generation RF amplifier design. The demonstrated reduction in second-order transconductance nonlinearity (g″m) confirms that latching contributes to a broader, flatter gm profile—a key requirement for signal linearity and spectral purity in communication systems. This gain in linearity, coupled with the absence of permanent degradation under prolonged high-field operation, makes SLCFETs a compelling platform for both military and commercial RF electronics. Moreover, the methodical combination of electroluminescent imaging, device modeling, and thermal analysis provided a blueprint for characterizing emerging effects in complex transistor structures. The ability to visualize and attribute localized emission to specific physical features—down to individual fins—adds a layer of diagnostic precision rarely seen in wide bandgap device research. This opens up possibilities for finer-grained control over switching dynamics in multichannel designs, particularly in environments where heat dissipation and voltage stress are limiting factors. Perhaps most critically, this study suggests that steep-slope switching in GaN need not rely on exotic materials or negative capacitance schemes. It can instead arise naturally from internal carrier dynamics when certain physical thresholds are crossed. That this transition is both recoverable and non-destructive makes it even more valuable. In doing so, the work not only advances the frontier of GaN transistor physics but also offers a realistic and manufacturable path to enhancing RF performance without compromising device longevity.

Lithography is the backbone of semiconductor and microelectromechanical system (MEMS) fabrication, where even the slightest deviation on the nanometer scale can ripple through an entire production chain, affecting both yield and reliability. As the push toward smaller device geometries intensifies, multi-exposure lithography has become indispensable for realizing ultra-fine patterning. However, this advancement brings its own complications—chief among them, the precision of alignment and the limited sensing range during overlay correction. These two factors now stand as central obstacles to the continued scaling of modern lithographic systems. Moiré fringe–based alignment techniques have, in many respects, been the preferred choice in the industry. Their appeal lies in their simple optical configuration, high sensitivity, and theoretical ability to achieve sub-nanometer resolution. Nonetheless, two deeply rooted issues constrain their full potential: the periodic aliasing of fringes and the spectral leakage encountered during phase extraction. The first limits the measurable displacement range to only a few micrometers, as the repeating periodicity of the Moiré pattern causes ambiguities once the misalignment exceeds roughly half of the grating period. The second arises during digital phase retrieval, where signal truncation and discrete sampling unavoidably distribute spectral energy into neighboring frequencies. Together, these effects compromise both range and precision, rendering traditional Moiré-based systems unsuitable for the next generation of nanofabrication tools that demand extended dynamic range without sacrificing accuracy. Researchers previously have explored various ways to overcome these limitations such as introducing static interferometric references, hierarchical grating designs, or beat-frequency modulation schemes and while each method brings a measure of improvement, most increase optical complexity or computation time, and few are robust enough for industrial-scale application. More recently, deep-learning-based algorithms have entered the scene, promising adaptive fringe interpretation but requiring extensive datasets and heavy processing power, which limit their practicality in real-time environments. Still, the most persistent challenge remains spectral leakage. In Moiré alignment imaging, the discrete Fourier transformation of non-integer sampled fringes inevitably spreads the spectral power, corrupting phase-unwrapping accuracy precisely where nanometer-level sensitivity is crucial. Containing this leakage without eroding spatial detail continues to be one of the more complex problems in precision lithography metrology. To this account, new research paper published in Optics Express and conducted by Feifan Xu, Jin Zhang, Weishi Li, and led by Professor Haojie Xia from the School of Instrument Science & Opto-electronics Engineering at Hefei University of Technology, The researchers developed two key innovations: a composite dual-frequency alignment mark consisting of upper and lower differential gratings, and an automatic lookup difference table (ALDT) algorithm for absolute misalignment computation. The dual-frequency mark enables interference patterns with extended periodicity, while the ALDT algorithm decodes large displacements unambiguously.

The research team designed a composite alignment mark in which the upper differential gratings (periods P₁ and P₂) functioned as measuring elements, while the lower gratings (P₃ and P₄) served as references. When illuminated by a coherent light source, interference between these grating pairs produced two sets of Moiré fringes—one with period PM₁₂ and the other with PM₃₄. Because these periods differ, the interference generates a dual-frequency pattern whose combined phase evolution encodes a unique mapping of wafer-mask misalignment. This arrangement effectively merges fine and coarse measurements into one optical process, allowing large-range detection without additional alignment marks. To extract accurate displacement data from the captured fringe images, the authors developed the ALDT algorithm. This method converts the offset of dual-frequency fringes into a set of arithmetic progressions, where each pair of upper and lower fringe shifts defines a unique misalignment interval. By establishing a lookup table of these intervals, the system can unambiguously determine the actual offset even when it exceeds the fundamental Moiré period. The result is a non-iterative, absolute measurement capable of spanning a range determined by the least common multiple (LCM) of the grating-derived base units. In practice, this translated to an expansion of the measurable range from 2.6 µm to 120 µm—a nearly 50-fold improvement over conventional systems.

The authors ensured phase accuracy and showed each fringe image processed with a 2D-HSCW prior to applying fast Fourier transform–based phase extraction. This self-convolution window function effectively suppressed spectral leakage that typically arises from non-integer truncation or camera sampling effects. Their simulations using MATLAB validated that the 2D-HSCW achieved unbiased phase estimation, even under significant image truncation. The precision of upper and lower fringe measurements remained within nanometer deviations across all test conditions, with the upper fringes consistently yielding superior accuracy. Then they validated experimentally on custom-fabricated alignment marks etched onto chrome-coated quartz wafers via ion-beam lithography. Using a collimated 530 nm LED source and a high-resolution CMOS camera, the researchers conducted step-displacement tests on a vibration-isolated optical platform. The ALDT algorithm maintained real-time performance, requiring only 0.15 seconds per measurement. Across multiple experimental runs, the system achieved sub-2-nm accuracy, with maximum errors below 2 nm and mean absolute errors under 1.54 nm.

In conclusion, the development reported by Professor Haojie Xia’s team represents a decisive step toward unifying high-precision and large-range lithography alignment within a single optical framework. Indeed, with the two-dimensional Hanning self-convolution window, the new system achieves sub-2-nm accuracy across a 120 µm range, offering a unified, real-time solution to lithography alignment challenges. The combination of composite dual-frequency Moiré marks and the ALDT algorithm resolves a long-standing tradeoff between measurement range and accuracy—a problem that has limited Moiré-based systems for decades. The researchers successfully established an elegant numerical scheme that bypasses periodic aliasing, allowing misalignment to be decoded uniquely across an extended range by interleaving the arithmetic intervals of two fringe sets,. This mathematical structure transforms what was once a relative measurement into an absolute one, delivering stability that is essential for the sub-nanometer domain. It is worth mentioning the role of the 2D-HSCW and by the authors addressing spectral leakage at the source rather than through post-hoc correction, they ensured the integrity of phase information even when imaging conditions deviate from ideal assumptions. The new method’s ability to achieve real-time processing while preserving nanometer accuracy demonstrates its suitability for in situ industrial integration. The fact that such precision was maintained across 120 µm of range marks a fundamental shift in how alignment sensors can be engineered for advanced semiconductor production. Beyond lithography, the framework opens new opportunities in precision metrology, MEMS calibration, optical displacement sensing, and nanoscale assembly. Any system requiring accurate relative position tracking—such as wafer bonding, deformable mirror alignment, or nanoscale robotics—can benefit from this dual-frequency Moiré strategy. Its algorithmic architecture, based on modular arithmetic and difference lookup, can also be adapted for three-dimensional positioning or even integrated with machine learning for self-correcting calibration. The implications extend further into manufacturing efficiency. Eliminating the need for separate coarse alignment steps simplifies the optical head design and shortens process cycles, reducing cost and complexity. As fabrication nodes approach the angstrom scale, where overlay tolerances become comparable to atomic dimensions, techniques capable of sub-2-nm reproducibility are indispensable. Professor Haojie Xia and colleagues approach therefore satisfies current industrial needs and also anticipates future requirements for ultra-dense integrated circuit production.

Depth-resolved wavelength-scanning interferometry (DRWSI) has become one of the most sensitive methods for measuring the surface shape and optical thickness of multilayer or transparent materials. Its capacity to reconstruct depth profiles from interferometric phase data enables precise characterization of micro-optical elements, thin films, and transparent plates. However, the classical implementation of DRWSI depends critically on the continuity of the laser’s wavelength scan. Diode lasers often have mode hopping or discontinuous spectral emission which produce gaps that disrupt the coherence of the interferometric signal. These discontinuities introduce random phase jitters and abnormal sidelobes that obscure the true topographic information and degrade the accuracy of phase recovery. Several approaches have previously been proposed to resolve this limitation and early strategies involved monitoring the wavelength-scanning sequence to reconstruct continuous spectra, but they inevitably narrowed the effective scanning range, reducing depth resolution. Other techniques, such as the random sampling Fourier transform (RSFT) and the iterative adaptive approach (IAA), reframed the problem as one of nonuniform spectral sampling or weighted least-squares fitting. However, although these methods succeeded in suppressing phase artifacts, they still relied on prior knowledge of the laser’s gapped spectrum—information not always available or measurable in real-time experiments. Furthermore, these algorithms typically demanded additional optical monitoring hardware, adding cost and complexity to what is otherwise an elegant and compact interferometric system. To this account, new research paper published in Optics Letters and conducted by Dr. Yulei Bai, Dr. Hao Qiu, Dr. Zean Huang, Dr. Zhaoshui He, Dr. Shengli Xie, and led by Professor Bo Dong from the School of Automation at Guangdong University of Technology, the researchers developed two complementary models: an auto-regression model to predict missing interference intensities within a gapped spectrum, and a discrete optimization model to automatically determine the number of missing points without prior spectral information.

The researchers first simulated DRWSI measurements under controlled spectral discontinuities to validate the ARM-based framework. They modeled the optical path difference using a MATLAB-generated surface shaped as the software’s logo and imposed gap intervals of 0.1 nm, 1 nm, and 10 nm within a 1.0-nm scanning range centered at 600 nm. The algorithm treated the missing intensities as unknowns within a vectorized interferometric signal, where an auto-regression of order γ estimated the missing components through least-squares prediction. A discrete optimization step then automatically determined the number of prediction points that minimized the total energy of the amplitude spectrum—including both mainlobe and sidelobes—after Fourier transformation. The authors found simulations produced compelling evidence of the method’s robustness. They reported that conventional Fourier transforms failed to resolve the true amplitude spectrum once gaps exceeded even 10 % of the scanning range, and generated multiple false interference peaks. On the other hand, the ARM reconstruction yielded a single, well-defined mainlobe consistent with the ideal theoretical spectrum. The resulting phase maps closely matched the ideal reference even when the discontinuity reached tenfold the nominal wavelength range, which confirmed that the model effectively predicted missing data without prior spectral information.

The authors afterward validated experimentally using a custom-built discontinuous DRWSI system incorporating a mode-hopping diode laser, a 4f optical setup, and a CCD detector capturing 308 frames during a 0.9-nm wavelength sweep. The sample combined a 6′ optical wedge and a USAF 1951 resolution target, providing four distinct reflective surfaces. Because the system lacked any hardware to measure the spectral gaps, neither RSFT nor IAA could be applied; only the conventional Fourier transform and the proposed ARM method were compared. The Fourier-based results revealed extensive spectral cross-talk and irregular sidelobes, which blurred the identification of true surface peaks and generated visible phase ripples. After applying the ARM model, these sidelobes were nearly eliminated, yielding smooth, high-fidelity phase distributions. Quantitatively, the wedge’s tilt angle derived from the ARM reconstruction deviated by only 0.35 % from the manufacturer’s specified 6′, a precision unmatched by conventional methods. Moreover, phase maps of the resolution target recovered clear structural patterns that had been completely obscured before correction.

In conclusion, the new study by Professor Bo Dong and colleagues successfully designed new models able to reconstruct continuous interference data from discontinuous wavelength scans. The new framework eliminates phase jitters and sidelobes caused by mode hopping and restored accurate depth-resolved phase maps while preserving full depth resolution. Indeed, it redefined the methodology of depth-resolved interferometry by decoupling spectral continuity from measurement fidelity. The auto-regression model introduces a self-learning mechanism into optical metrology: instead of compensating for hardware imperfections through calibration, the signal itself guides the recovery of lost spectral information. The method also achieves a level of adaptability unattainable in traditional Fourier-based frameworks by automatically determining the missing points through discrete optimization. It restores the core advantages of DRWSI—full-field, high-sensitivity phase detection—while lifting the stringent requirement for continuous laser emission. Additionally, the newly proposed technique broadens the scope of DRWSI applications. Compact diode lasers, often dismissed for precision metrology due to mode-hopping instabilities, can now be integrated into low-cost, high-accuracy instruments. This advance could benefit optical component manufacturing, multilayer film inspection, and biomedical surface imaging, where system simplicity and flexibility are valued alongside precision. Moreover, because the algorithm operates entirely in software and requires no auxiliary spectral-monitoring device, it reduces maintenance and system calibration time.

The new study challenges the assumption that continuous spectral scanning is indispensable for accurate phase retrieval and with their demonstration that an unknown gapped spectrum can yield equally reliable topographic information, it invites a conceptual shift: precision need not rely solely on hardware perfection but can emerge from intelligent modeling. Moreover, the proposed method effectively overcomes the limitation of phase demodulation accuracy caused by wavelength discontinuity in multi-source swept-frequency tomographic imaging. In a nutshell, the new findings demonstrated that the ARM approach restored phase accuracy, achieving performance previously possible only with continuous-spectrum lasers. Future studies might extend this concept to hyperspectral interferometry, optical coherence tomography, or spectroscopic imaging, where spectral gaps hinder data fidelity.