Electronic Unit Research Articles

Advancing interactive systems with liquid crystal network-based adaptive electronics.

Achieving adaptive behavior in artificial systems, analogous to living organisms, has been a long-standing goal in electronics and materials science. Efforts to integrate adaptive capabilities into synthetic electronics traditionally involved a typical architecture comprising of sensors, an external controller, and actuators constructed from multiple materials. However, challenges arise when attempting to unite these three components into a single entity capable of independently coping with dynamic environments. Here, we unveil an adaptive electronic unit based on a liquid crystal polymer that seamlessly incorporates sensing, signal processing, and actuating functionalities. The polymer forms a film that undergoes anisotropic deformations when exposed to a minor heat pulse generated by human touch. We integrate this property into an electric circuit to facilitate switching. We showcase the concept by creating an interactive system that features distributed information processing including feedback loops and enabling cascading signal transmission across multiple adaptive units. This system responds progressively, in a multi-layered cascade to a dynamic change in its environment. The incorporation of adaptive capabilities into a single piece of responsive material holds immense potential for expediting progress in next-generation flexible electronics, soft robotics, and swarm intelligence.

Development and Validation of an Explainable Deep Learning Model to Predict In-Hospital Mortality for Patients With Acute Myocardial Infarction: Algorithm Development and Validation Study.

Acute myocardial infarction (AMI) is one of the most severe cardiovascular diseases and is associated with a high risk of in-hospital mortality. However, the current deep learning models for in-hospital mortality prediction lack interpretability. This study aims to establish an explainable deep learning model to provide individualized in-hospital mortality prediction and risk factor assessment for patients with AMI. In this retrospective multicenter study, we used data for consecutive patients hospitalized with AMI from the Chongqing University Central Hospital between July 2016 and December 2022 and the Electronic Intensive Care Unit Collaborative Research Database. These patients were randomly divided into training (7668/10,955, 70%) and internal test (3287/10,955, 30%) data sets. In addition, data of patients with AMI from the Medical Information Mart for Intensive Care database were used for external validation. Deep learning models were used to predict in-hospital mortality in patients with AMI, and they were compared with linear and tree-based models. The Shapley Additive Explanations method was used to explain the model with the highest area under the receiver operating characteristic curve in both the internal test and external validation data sets to quantify and visualize the features that drive predictions. A total of 10,955 patients with AMI who were admitted to Chongqing University Central Hospital or included in the Electronic Intensive Care Unit Collaborative Research Database were randomly divided into a training data set of 7668 (70%) patients and an internal test data set of 3287 (30%) patients. A total of 9355 patients from the Medical Information Mart for Intensive Care database were included for independent external validation. In-hospital mortality occurred in 8.74% (670/7668), 8.73% (287/3287), and 9.12% (853/9355) of the patients in the training, internal test, and external validation cohorts, respectively. The Self-Attention and Intersample Attention Transformer model performed best in both the internal test data set and the external validation data set among the 9 prediction models, with the highest area under the receiver operating characteristic curve of 0.86 (95% CI 0.84-0.88) and 0.85 (95% CI 0.84-0.87), respectively. Older age, high heart rate, and low body temperature were the 3 most important predictors of increased mortality, according to the explanations of the Self-Attention and Intersample Attention Transformer model. The explainable deep learning model that we developed could provide estimates of mortality and visual contribution of the features to the prediction for a patient with AMI. The explanations suggested that older age, unstable vital signs, and metabolic disorders may increase the risk of mortality in patients with AMI.

Self-Diagnostic Opportunities for Battery Systems in Electric and Hybrid Vehicles

The number of battery systems is also growing significantly along with the rise in electric and hybrid car sales. Different vehicles use different types and numbers of batteries. Furthermore, the layout and operation of the control and protection electronics units may also differ. The research aims to develop an approach that can autonomously detect and localize the weakest cells. The method was validated by testing the battery systems of three different VW e-Golf electric vehicles. A wide-range discharge test was performed to examine the condition assessment and select the appropriate state of charge (SoC) for all three vehicles. On the one hand, the analysis investigated the cell voltage deviations from the average; the tests cover deviations of 0 mV, 12 mV, 60 mV, 120 mV, and 240 mV. On the other hand, the mean value calculation was used to filter out possible erroneous values. Another important aspect was examining the relationship between the state of charges (SoC) and the deviations. Therefore, the 10% step changes were tested to see which SoC level exhibited more significant voltage deviations. Based on the results, it was observed that there are differences between the cases, and the critical range is not necessarily at the lowest SoC level. Furthermore, the load rate (current) and time of its occurrence play an important role in the search for a faulty cell. An additional advantage of this approach is that the process currently being tested on the VW e-Golf can be relatively simply transferred to other types of vehicles. It can also be a very useful addition for autonomous vehicles, as it can self-test the cells in the system at low power consumption.

CFD simulation and geometric optimization of a novel baffle heat sink for power electronics cooling

The electric drivetrain of an electric vehicle typically has two fluid circuits: an oil circuit for lubrication and a water-glycol circuit for cooling. Eliminating the water-glycol circuit can result in a more compact and lighter drivetrain. However, this requires the drivetrain components to be cooled with oil as coolant, which has inferior heat transfer properties compared to water-glycol mixtures. This is particularly challenging for the power modules in the power electronics unit, which exhibit some of the highest heat fluxes in the drivetrain. This study focusses on the design and optimization of a novel type of heat sink for laminar flows with baffles to guide the flow in four different directions. This has two main advantages: the baffles act as fins and increase the heat transfer area and they introduce a swirling motion in the oil thereby disturbing the boundary layers continuously. Computational fluid dynamics simulations using Ansys Fluent are performed to evaluate the thermohydraulic performance of the heat sink. The influence of design parameters such as the height and thickness of both fins and baffles and the baffle spacing in horizontal, upward and downward direction was evaluated by simulations with several combinations of the design space, considering manufacturing constraints. The optimization of the design showed that the fin thickness, horizontal baffle spacing and baffle length should be as small as possible, while there were optimal values for fin height, upward and downward baffle spacing, streamwise baffle spacing and baffle thickness. The simulations show that the thermal resistance of the baffle heat sink can be 19% lower than that of a benchmark pin fin heat sink at equal pumping power.

Molecular-Potential and Redox Coregulated Cathodic Electrosynthesis toward Ionic Azulene-Based Thin Films for Organic Memristors.

Organic memristors as promising electronic units are attracting significant attention owing to their simplicity of molecular structure design. However, fabricating high-quality organic films via novel synthetic technologies and exploring unprecedented chemical structures to achieve excellent memory performance in organic memristor devices are highly challenging. In this work, we report a cathodic electropolymerization to synthesize an ionic azulene-based memristive film (PPMAz-Py+Br-) under the molecular-potential and redox coregulation. During the cathodic electropolymerization process, electropositive pyridinium salts migrate to the cathode under an electric field, undergo a reduction-coupling deprotonation reaction, and polymerize into a uniform film with a controllable thickness on the electrode surface. The prepared Al/PPMAz-Py+Br-/ITO devices not only exhibit a high ON/OFF ratio of 1.8 × 103, high stability, long memory retention, and endurance under a wide range of voltage scans, but also achieve excellent multilevel storage and history-dependent memristive performance. In addition, the devices can mimic important biosynaptic functions, such as learning/forgetting function, synaptic enhancement/inhibition, paired-pulse facilitation/depression, and spiking-rate-dependent plasticity. The tunable memristive performances are attributed to the capture of free electrons on pyridinium cations, the migration of the aluminum ions (Al3+), and the form of Al conductive filaments under voltage scans.

Diagnosing and characterizing inflammatory myopathies at an Australian tertiary public hospital: Resource utilization and direct healthcare costs.

To determine the direct health service costs and resource utilization associated with diagnosing and characterizing idiopathic inflammatory myopathies (IIMs), and to assess for limitations and diagnostic delay in current practice. A retrospective, single-center cohort analysis of all patients diagnosed with IIMs between January 2012 and December 2021 in a large tertiary public hospital was conducted. Demographics, resource utilization and costs associated with diagnosing IIM and characterizing disease manifestations were identified using the hospital's electronic medical record and Health Intelligence Unit, and the Medicare Benefits Schedule. Thirty-eight IIM patients were identified. IIM subtypes included dermatomyositis (34.2%), inclusion body myositis (18.4%), immune-mediated necrotizing myopathy (18.4%), polymyositis (15.8%), and anti-synthetase syndrome (13.2%). The median time from symptom onset to diagnosis was 212 days (IQR: 118-722), while the median time from hospital presentation to diagnosis was 30 days (8-120). Seventy-six percent of patients required emergent hospitalization during their diagnosis, with a median length of stay of 8 days (4-15). The average total cost of diagnosing IIM was $15 618 AUD (STD: 11331) per patient. Fifty percent of patients underwent both MRI and EMG to identify affected muscles, 10% underwent both pan-CT and PET-CT for malignancy detection, and 5% underwent both open surgical and percutaneous muscle biopsies. Autoimmune serology was unnecessarily repeated in 37% of patients. The diagnosis of IIMs requires substantial and costly resource use; however, our study has identified potential limitations in current practice and highlighted the need for streamlined diagnostic algorithms to improve patient outcomes and reduce healthcare-related economic burden.

Building a Fully-Automatized Active Learning Framework for the Semantic Segmentation of Geospatial 3D Point Clouds

In recent years, significant progress has been made in developing supervised Machine Learning (ML) systems like Convolutional Neural Networks. However, it’s crucial to recognize that the performance of these systems heavily relies on the quality of labeled training data. To address this, we propose a shift in focus towards developing sustainable methods of acquiring such data instead of solely building new classifiers in the ever-evolving ML field. Specifically, in the geospatial domain, the process of generating training data for ML systems has been largely neglected in research. Traditionally, experts have been burdened with the laborious task of labeling, which is not only time-consuming but also inefficient. In our system for the semantic interpretation of Airborne Laser Scanning point clouds, we break with this convention and completely remove labeling obligations from domain experts who have completed special training in geosciences and instead adopt a hybrid intelligence approach. This involves active and iterative collaboration between the ML model and humans through Active Learning, which identifies the most critical samples justifying manual inspection. Only these samples (typically ≪1%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\ll 1{\\%}$$\\end{document} of Passive Learning training points) are subject to human annotation. To carry out this annotation, we choose to outsource the task to a large group of non-specialists, referred to as the crowd, which comes with the inherent challenge of guiding those inexperienced annotators (i.e., “short-term employees”) to still produce labels of sufficient quality. However, we acknowledge that attracting enough volunteers for crowdsourcing campaigns can be challenging due to the tedious nature of labeling tasks. To address this, we propose employing paid crowdsourcing and providing monetary incentives to crowdworkers. This approach ensures access to a vast pool of prospective workers through respective platforms, ensuring timely completion of jobs. Effectively, crowdworkers become human processing units in our hybrid intelligence system mirroring the functionality of electronic processing units.

Frequency-comb-linearized, widely tunable lasers for coherent ranging

Tunable lasers, with the ability to continuously vary their emission wavelengths, have found widespread applications across various fields such as biomedical imaging, coherent ranging, optical communications, and spectroscopy. In these applications, a wide chirp range is advantageous for large spectral coverage and high frequency resolution. Besides, the frequency accuracy and precision also depend critically on the chirp linearity of the laser. While extensive efforts have been made on the development of many kinds of frequency-agile, widely tunable, narrow-linewidth lasers, wideband yet precise methods to characterize and linearize laser chirp dynamics are also demanded. Here we present an approach to characterize laser chirp dynamics using an optical frequency comb. The instantaneous laser frequency is tracked over terahertz bandwidth at 1 MHz intervals. Using this approach we calibrate the chirp performance of 12 tunable lasers from Toptica, Santec, New Focus, EXFO, and NKT that are commonly used in fiber optics and integrated photonics. In addition, with acquired knowledge of laser chirp dynamics, we demonstrate a simple frequency-linearization scheme that enables coherent ranging without any optical or electronic linearization unit. Our approach not only presents novel wideband, high-resolution laser spectroscopy, but is also critical for sensing applications with ever-increasing requirements on performance.

First- and second-order phase transitions in electronic excitable units and neural dynamics under global inhibitory feedback

The diverse roles of inhibition in neural circuits and other dynamical networks are receiving renewed interest. Here, it is shown that increasing global inhibitory feedback leads to gradual rounding of first-order transition between dynamical phases, turning it into second-order transition. The effect is initially observed in an electronic model consisting of a bi-dimensional array of neon glow lamps, where global inhibition can be simply introduced through a resistor in series with the supply voltage. The experimental findings are confirmed using both an extended numerical model and a mean-field approximation, then replicated across different models of neural dynamics, namely, the Wilson–Cowan model and a network of leaky integrate-and-fire neurons. Across all these systems, a critical point is always found as a function of a pair of parameters controlling local excitability and global inhibition strength, and a general explanation revealing the roles of the shape of the activation function and voltage fluctuations versus the extinction time-scale is provided. It is speculated that the brain could use global inhibition as a versatile means of shifting between first- and second-order dynamics, addressing the conundrum regarding the coexistence in neural dynamics of phenomena stemming from both. Some reflections regarding the comparison with other physical systems and the possible physiological significance are offered, and a hypothetical setup for an optogenetics experiment on cultured neurons is put forward.

Research of the vacuum brake booster working process

The creation of a laboratory installation was carried out with an aim to ensure the study of the working processes and characteristics of vacuum boosters as a part of hydraulic brake drives for vehicles with a gross weight of up to 3.5 tons for civilian usage, and armored vehicles with a gross weight of up to 8.5 tons designated for service and combat missions. Theoretical researches in this direction have been previously carried out by a number of scientists of the Department of “Automobiles” named after prof. Gredeskul A.B. in Kharkiv National Automobile and Road University, the results of which have been highlighted in a number of scientific papers. Comparison of the results of theoretical studies with the experimental ones, received on the laboratory setup and suggested in this paper, according to the experimental method described in the article, using the created electronic signal processing complex and sensor unit, made it possible to establish a discrepancy between theoretical and experimental studies within 6%. This complex for experimental research was created for the first time, thus allowing to obtain the results that confirm the theoretical studies of the vacuum boosters of the brake drive of cars, as well as revealed a number of dependencies between the weight and overall parameters of the under research unit, along with the number of functional relationships between the structural components of the vacuum boosters, which enables to significantly optimize its design.

Integrating superregenerative principles in a compact, power-efficient NMR/NQR spectrometer: A novel approach with pulsed excitation

We present a new approach to Nuclear Quadrupole Resonance (NQR)/Nuclear Magnetic Resonance (NMR) spectroscopy, the Damp-Enhanced Superregenerative Nuclear Spin Analyser (DESSA). This system integrates Superregenerative principles with pulsed sample excitation and detection, offering significant advancements over traditional Super-Regenerative Receivers (SRRs). Our approach overcomes certain limitations associated with traditional Super-Regenerative Receivers (SRRs) by integrating direct digital processing of the oscillator response delay time (Td) and an electronic damp unit to regulate the excitation pulse decay time (Te). The essence is combining pulsed excitation with a reception inspired by, but distinct from, conventional SRRs. The damp unit allows a rapid termination of the oscillation pulse and the initiation of detection within microseconds, and direct digital processing avoids the need for a second lower frequency which is used for quenching in a traditional SRRs, thereby avoiding the formation of sidebands. We demonstrate the effectiveness of DESSA on a NaClO3 sample containing the isotope Chlorine-35 where it accurately detects the NQR signal with sub-kHz resolution.

An on-chip microarray platform for material-temperature optimization and gas discrimination

The demand for gases detection has promoted extensive research into gas sensors. Sensor arrays can be used as electronic olfaction units for complex gas detection and identification. However, the gas sensing properties are affected by various factors. The multi-dimensional nature of those interactions is crucial for creating a sensor array with diverse responses, but it is also challenging to fabricate an array considering all these parameters, particularly in miniaturization. In this paper, we present a framework that can take multiple factors into consideration and help to fabricate an optimal array for gas identification. 32 sensors including different MOS materials (SnO2, In2O3, WO3), morphologies (microsphere, nanoparticle, nanoflake), doped/loaded metals (Pd, Pt, Ce, Ru, Rb), and operational temperatures (250℃∼370℃) are discussed. A material-temperature-analyte-response database was then established by storing their response characteristics towards low ppm concentrations of ammonia, ethyl acetate, toluene, methanol, ethanol, and acetone. An optimized array was then constructed out of 10,518,300 possible combinations of materials and operational temperatures. Results showed improvements in discrimination ability and the optimized array integrated into the chip can provide vector signals with high selectivity, completely recognizing all the 6 target gases and even for their multi-component mixture after linear discriminant analysis transformation.

Synthesis of Electronic Frequency Tuning Units for Narrowband Ultra-Low Noise Oscillators with Surface Acoustic Waves Resonators

Synthesis of Electronic Frequency Tuning Units for Narrowband Ultra-Low Noise Oscillators with Surface Acoustic Waves Resonators

Examination of a Magnetorheological Damper Control System with Vibration Energy Harvesting

Abstract The study deals with the experimental examination of a magnetorheological (MR) damper control system with vibration energy harvesting using a specially engineered electronic unit (EU). Unlike a typical MR damper control system, which requires an external energy source, the developed system is powered exclusively by energy extracted from a vibrating structure (mechanical system with one-degree-of freedom) and processed through the EU. The work describes the structure and functions of the EU, presents the test rig and the control algorithm implementation, and discusses the test results of the control system under harmonic kinematic excitations of low frequency range.

Aircraft vulnerability to politically motivated Radio Frequency Interference (RFI) in Eastern Europe

Countries in Europe have experienced radio frequency interference from Russian electronic warfare units, affecting navigation systems without discrimination. Interference has been identified as coming from the Russian mainland, Kaliningrad, and in the areas where it is engaged in conflicts abroad, creating serious hazards for aircraft. Previous research has identified the technical vulnerabilities to aviation from radio frequency interference, but it has yet to be contextualised in light of the ongoing geopolitical tensions. Using literature review analysis, the aim of the article is to place the jamming and spoofing threat in the context of ongoing political tensions between Russia and the region and to establish worse-case scenarios based on the former’s motives. Focusing on the threats to aircraft, it finds that the likely motivations are to (1) complement political narratives of Western aggression; (2) to deny service for intimidation, harassment, economic loss, and to portray a dominant Russian cyber influence; and (3) to use spoofing to enable hostage diplomacy via the seizure of people and assets. It argues that reporting mechanisms for aviation risks assumes geographical staticism, which does not correspond adequately to the threat. It also creates arbitrary “predictability” in otherwise unpredictable environments, which could ultimately affect vigilance and due diligence in the areas not considered problematic.

Fibrous MXene Synapse-Based Biomimetic Tactile Nervous System for Multimodal Perception and Memory.

Biomimetic tactile nervous system (BTNS) inspired by organisms has motivated extensive attention in wearable fields due to its biological similarity, low power consumption, and perception-memory integration. Though many works about planar-shape BTNS are developed, few researches could be found in the field of fibrous BTNS (FBTNS) which is superior in terms of strong flexibility, weavability, and high-density integration. Herein, a FBTNS with multimodal sensibility and memory is proposed, by fusing the fibrous poly lactic acid (PLA)/Ag/MXene/Pt artificial synapse and MXene/EMIMBF4 ionic conductive elastomer. The proposed FBTNS can successfully perceive external stimuli and generate synaptic responses. It also exhibits a short response time (23ms) and low set power consumption (17 nW). Additionally, the proposed device demonstrates outstanding synaptic plasticity under both mechanical and electrical stimuli, which can simulate the memory function. Simultaneously, the fibrous devices are embedded into textiles to construct tactile arrays, by which biomimetic tactile perception and temporary memory functions are successfully implemented. This work demonstrates the as-prepared FBTNS can generate biomimetic synaptic signals to serve as artificial feeling signals, it is thought that it could offer a fabric electronic unit integrating with perception and memory for Human-Computer interaction, and has great potential to build lightweight and comfortable Brain-Computer interfaces.

Acute kidney injury comorbidity analysis based on international classification of diseases-10 codes.

Acute kidney injury (AKI) is a clinical syndrome that occurs as a result of a dramatic decline in kidney function caused by a variety of etiological factors. Its main biomarkers, serum creatinine and urine output, are not effective in diagnosing early AKI. For this reason, this study provides insight into this syndrome by exploring the comorbidities of AKI, which may facilitate the early diagnosis of AKI. In addition, organ crosstalk in AKI was systematically explored based on comorbidities to obtain clinically reliable results. We collected data from the Medical Information Mart for Intensive Care-IV database on patients aged [Formula: see text] 18 years in intensive care units (ICU) who were diagnosed with AKI using the criteria proposed by Kidney Disease: Improving Global Outcomes. The Apriori algorithm was used to mine association rules on the diagnoses of 55,486 AKI and non-AKI patients in the ICU. The comorbidities of AKI mined were validated through the Electronic Intensive Care Unit database, the Colombian Open Health Database, and medical literature, after which comorbidity results were visualized using a disease network. Finally, organ diseases were identified and classified from comorbidities to investigate renal crosstalk with other distant organs in AKI. We found 579 AKI comorbidities, and the main ones were disorders of lipoprotein metabolism, essential hypertension, and disorders of fluid, electrolyte, and acid-base balance. Of the 579 comorbidities, 554 were verifiable and 25 were new and not previously reported. In addition, crosstalk between the kidneys and distant non-renal organs including the liver, heart, brain, lungs, and gut was observed in AKI with the strongest heart-kidney crosstalk, followed by lung-kidney crosstalk. The comorbidities mined in this study using association rules are scientific and may be used for the early diagnosis of AKI and the construction of AKI predictive models. Furthermore, the organ crosstalk results obtained through comorbidities may provide supporting information for the management of short- and long-term treatment practices for organ dysfunction.

Reliability Thermal Design, Simulation and Experimental Verification of Electronic Monitoring Unit for Magnetic Resonance System

The monitoring unit used in the nuclear magnetic resonance system, as an important unit of the system, faces a high thermal risk during its entire life cycle. This paper ensures the high efficiency and reliability of the thermal design of the product module from the two dimensions of structural design and device derating design. In order to reduce the risk of thermal design of electronic modules and comprehensively verify the effectiveness of thermal design of electronic modules, the design verification is carried out by combining simulation and experiment. In the simulation process, by establishing a thermal simulation model at the circuit board level, the crustal temperature of the core device is numerically calculated, and the index is compared with the thermal design index value and the test value, on the one hand, to verify the correctness of the simulation model. On the other hand, the validity of thermal design is verified. In the testing process, a thermal test platform for product modules is built, and the thermal characteristics test values of the core components of the module under extreme electrical conditions are obtained, and the corresponding conversion methods are used to predict the thermal performance and thermal design margin of the product at different altitudes. The results show that the electronic module can meet the thermal design requirements in terms of structural design and derating design of core components, and can ensure that the product module can work safely and reliably during the entire life cycle of the NMR system.

NDE of Concrete Strength and Elasticity Modulus Based on Jerk Measurement

The quality assessment of concrete material for new or existing buildings is critical to civil and structural engineers. In this scope, the development of nondestructive testing methods (NDT) has received significant attention and becomes essential for enhancing the on-site assessment of concrete performance. The Schmidt hammer (SH) test is one of the most widely used NDT for quantifying concrete strength. However, the reliability of this test and its accuracy are questioned by many researchers. In this study, the measurement quality of SH has been improved by attaching a developed cantilever-based jerk sensor to the SH mass. The theoretical model of the proposed measurement technique has been formulated and discussed. The experimental approach, jerk sensor design, SH body modification, electronic reading unit, testing, and construction of the calibration curves have been presented. Experimental results revealed that the accuracy of the developed SH has increased by 13% for concrete strength measurement, and the correlation coefficient of the strength calibration curve is 0.94. At the same time, the calibration curve of the concrete elasticity modulus showed a 0.93 correlation coefficient with 92% measurement accuracy. Analytical and experimental results have confirmed the applicability of the modified Schmidt hammer since it is reliable and more accurate compared to the traditional Schmidt Hammer.

M3ICRO: Machine learning-enabled compact photonic tensor core based on programmable multi-operand multimode interference

Photonic computing shows promise for transformative advancements in machine learning (ML) acceleration, offering ultrafast speed, massive parallelism, and high energy efficiency. However, current photonic tensor core (PTC) designs based on standard optical components hinder scalability and compute density due to their large spatial footprint. To address this, we propose an ultracompact PTC using customized programmable multi-operand multimode interference (MOMMI) devices, named M3ICRO. The programmable MOMMI leverages the intrinsic light propagation principle, providing a single-device programmable matrix unit beyond the conventional computing paradigm of one multiply-accumulate operation per device. To overcome the optimization difficulty of customized devices that often requires time-consuming simulation, we apply ML for optics to predict the device behavior and enable differentiable optimization flow. We thoroughly investigate the reconfigurability and matrix expressivity of our customized PTC and introduce a novel block unfolding method to fully exploit the computing capabilities of a complex-valued PTC for near-universal real-valued linear transformations. Extensive evaluations demonstrate that M3ICRO achieves a 3.5–8.9× smaller footprint, 1.6–4.4× higher speed, 9.9–38.5× higher compute density, 3.7–12× higher system throughput, and superior noise robustness compared to state-of-the-art coherent PTC designs. It also outperforms electronic digital A100 graphics processing unit by 34.8–403× higher throughput while maintaining close-to-digital task accuracy across various ML benchmarks.