Efficient and automated wafer yield prediction is central to cost control and process optimization in intelligent semiconductor manufacturing. As the initial testing stage in wafer inspection, Wafer Acceptance Testing (WAT) data contain critical process-related information. However, its high dimensionality, redundancy, and nonlinear inter-dependencies pose significant challenges to conventional yield prediction models, such as high computational overhead and limited generalization capability. Moreover, existing approaches often lack an automated framework capable of jointly addressing feature redundancy and model complexity. This paper proposes a collaborative and automated prediction framework that integrates feature selection and machine learning. First, an enhanced Binary Zebra Population Optimization Algorithm (BZPOA) is introduced, which incorporates redesigned exploration and development mechanisms to automatically identify key feature subsets from high-dimensional parameters, substantially reducing data redundancy and computational dimensionality. Second, a Bayesian hyperparameter-optimized XGBoost model is constructed, utilizing the Tree-structured Parzen Estimator (TPE) to achieve deep co-optimization of model parameters and feature space, thereby overcoming the inefficiency and overfitting issues commonly associated with manual parameter tuning. Experiments on a real-world dataset demonstrate that the proposed framework achieves average Recall, Precision, and F1-scores of 0.872, 0.917, and 0.902, respectively. Compared with the full-feature baseline, the BZPOA-selected feature subset improves predictive performance by 5%–10%, attains an AUC of 0.904, and significantly reduces per-wafer prediction time. Cross-factory transfer experiments further confirm the robustness of the proposed system.

Logic-level defects that escape manufacturing tests pose reliability risks in modern systems and require functional testing to identify their activation and propagation behaviors. However, effective functional testing is limited by the high cost of long-cycle fault simulations. To address this challenge, we propose a Spatio-Temporal Graph Convolutional Network framework to efficiently and accurately predict the Fault Impact Probability on the circuit's function cross-over multiple function cycles, enabling rapid quantitative assessment of functionally possible faults. Our method represents gate-level netlists as spatio-temporal graphs, capturing both structural connectivity and short-range signal-propagation dynamics. With dedicated spatial and temporal encoders, the proposed ST-GCN enables accurate prediction of multi-cycle circuit-level FIP. Experiments on ISCAS’89 benchmarks show that the approach reduces fault-simulation cost by over an order of magnitude while maintaining high accuracy (mean absolute error as low as 0.024 for 5-cycle predictions). The framework supports both testability-metric-based and simulation-based feature construction, enabling a tunable balance between efficiency and accuracy. A case study on test point selection further demonstrates that using predicted FIPs to guide observation-point placement improves the detectability of multi-cycle, hard-to-detect circuit-level faults. Overall, this work provides a scalable solution for circuit-level multi-cycle fault-impact assessment and can be readily integrated into functional test generation and other Electronic Design Automation workflows.

Chiplet systems that stack heterogeneous dies via 2.5D/3D integration use JTAG-based test access ports (TAPs) to validate inter-die links and enable in-field diagnosis. However, these TAPs create a shared attack surface: penetrating a single die can potentially expose control of the entire stack. Current countermeasures involve embedding a complete cryptographic engine in each chiplet, which increases the area and locks the protocol at tape-out, leaving it susceptible to future unknown attacks. This paper proposes a Software-Defined Secure Island (SDSI) architecture that decouples security policies from hardwired logic while satisfying non-functional requirements. Each chiplet instantiates an ultra-lightweight Secure Island Controller (SIC) macro that performs only two XORs and two additions per handshake. Meanwhile, a centralized secure island runs heavyweight cryptography in firmware using the multi-round SASL-JTAG+ protocol. SDSI enables scalable, adaptable test access protection by decoupling security from hardware. An FPGA implementation shows that the SIC macro occupies 11% of the area of an AES- 128 core and 37% of a SHA256 core, yet it supports 256 - to 512-bit keys with negligible growth. A security analysis demonstrates immunity to replay attacks because the authentication data is refreshed with each session. All future upgrades, such as longer keys, stronger hashes, and additional rounds, are delivered via firmware, providing scalable, field-upgradable protection for heterogeneous chiplet systems.

Although advanced 3D chiplet packaging helps extend Moore’s-Law gains through vertical stacking of heterogeneous dies, it simultaneously introduces unprecedented thermal challenges. Shrinking 3D interconnects and rising current densities trap heat at through-silicon vias (TSVs) and micro-bumps, causing steep temperature gradients and thermo-mechanical stress that directly trigger early failures. Existing methods interact with finite-element analysis (FEA) tools mainly through manual, experience-based parameter tuning, inevitably making hot-spot detection time-consuming and error-prone. This paper establishes a text-interactive, thermal-failure-aware design framework for 3D chiplets that couples large language model (LLM)-accelerated thermal analysis with FEA simulation. A Python-driven Automatic Code Generation Model (ACGM) is invoked to automate geometry drawing, meshing, and boundary-condition assignment, enabling rapid analysis and prediction of hot-spot locations within densely stacked chiplets and TSVs. The proposed ACGM eliminates domain-expert dependence by abstracting cumbersome FEA parameter setting into natural-language commands, thereby lowering the entry barrier and cutting operation time. Deployment of the proposed ACGM streamlines intricate FEA operations and Python scripting into concise natural-language commands, enabling accurate layout-parameter tuning and shortened design cy-cles-establishing an AI-centric design-verification paradigm for advanced 3D chiplets.

Soft errors have been a severe threat to the reliability of modern integrated circuits (ICs), making hardened latch designs indispensable for masking soft errors with redundancy. However, the added redundancy also masks production defects as soft errors; this makes it hard to detect defects in hardened latches, thus significantly reducing their reliability. Our previous work proposed the scan-test-aware hardened latch (STAHL) design, the first for addressing the issue of low defect detectability of hardened latch designs. However, STAHL still suffers from two problems: 1) it is not self-resilient to soft errors and 2) a STAHL-based scan design requires one additional control signal. This article proposes a high defect detectable and single-event-upset (SEU)-resilient robust (HIDER) latch to address the issues of the low defect detectability of existing hardened latches and the STAHLs lack of SEU-resilient capability. Two scan designs [HIDER-based scan-cell-S (HIDER-SC-S) and HIDER-based scan-cell-F (HIDER-SC-F)], as well as two corresponding test procedures, are proposed to fully test HIDER latch with only one control signal. Simulation results show that the HIDER latch achieves the highest defect coverage (DC) in both single latch cell detection and scan tests among all existing hardened latch designs. In addition, the HIDER latch has much lower power and a smaller delay than STAHL.

As a critical hardware security primitive for the authentication within the Internet of Things (IoT), the physical unclonable function (PUF) represents an innovative security design paradigm for integrated circuits. However, the linear challenge-response mapping of the arbiter PUF (APUF) and variants render these structures more susceptible to modeling attacks due to their delayed linear structure. In this article, we propose a nonlinearized demultiplexer time-to-digital converter (DEMUX-TDC) PUF. This PUF uses a quantized delay difference technique to alter the traditional response generation mechanism, demonstrating robust resistance against modeling attacks. First, the proposed PUF employs a segmented APUF variant structure configured in both front and back segmented modes to generate a source of delay difference entropy. Additionally, the scheme incorporates a multilayer differential tapped TDC circuit to quantize the delay differences into digital codes, followed by a linear feedback shift register (LFSR) to obfuscate the final output response. We further propose a highly secure mutual authentication protocol based on reconfigurable PUF, by leveraging the characteristics of front and back segments of the PUF’s challenge. Evaluation of the proposed scheme on the implementation of Xilinx Virtex-7 and Spartan-6 field-programmable gate array (FPGA) demonstrates that the uniqueness and uniformity can reach ideal value, in the condition of prediction accuracy across six modeling attacks remaining around 50%.

This paper proposes a novel Test Point Insertion (TPI) strategy to enhance the testability for multi-cycle Built-In Self-Test (BIST) for logic circuits. The approach leverages Multivariate Temporal-Spatial Graph Convolutional Neural Networks (MTS-GCN) and Reinforcement Learning to identify optimal Test Points (TPs). The proposed TPI method treats the testability information of a logic circuit as time-series data and employs Multivariate Time-Series Graph Neural Networks (MTGNN) to capture the relationship between the circuit's structural (spatial information) attributes and the temporal variability of signal line testability across capture cycles. A subsequent Multi-Layer Perceptron (MLP) computes the metric for each signal line to pinpoint potential TPs based on the extracted temporal-spatial features. Experimental evaluation based on benchmark circuits confirms the efficacy of the proposed model, which is trained with Deep Q-Networks (DQN), in improving the fault detection for multi-cycle logic BIST.

True random number generators (TRNGs), as an important component of security systems, have received a lot of attention for their related research. The previous researches have provided a large number of TRNG solutions, however, they still failed to reach an excellent tradeoff in various performance metrics. This article presents a shift-registers metastability-based TRNG, which is implemented by compact reference units and comparison units. By forcing the D flip-flops in the shift-registers into the metastable state, it optimizes the problem that the conventional metastability entropy sources consume excessive hardware resources. And a new method of metastable randomness extraction is used to reduce the bias of metastable output. The proposed TRNG is implemented in Xilinx Spartan-6 and Virtex-6 FPGAs, which generate random sequences that pass the NIST SP800-22, NIST SP800-90B tests and show excellent robustness to voltage and temperature variations. This TRNG can consume only 3 slices of the FPGA, but it has a high throughput rate of 25 Mbit/s. In comparison with state-of-the-art FPGA-compatible TRNGs, the proposed TRNG achieves the highest figure of merit FOM, which means that the proposed TRNG significantly outperforms previous researches in terms of hardware resources, throughput rate, and operating frequency tradeoffs.

Defect diagnosis and localization in wafer maps are crucial tasks in semiconductor manufacturing. Existing deep learning methods often require pixel-level annotations, making them impractical for large-scale deployment. In this paper, we propose a novel weakly supervised learning approach to achieving high-precision defect identification and effective localization with only image-level labels. By leveraging the information of defect types and locations, we introduce a weighted fusion of activation maps, called Class Activation Map (CAM), to highlight classspecific regions. We further enhance defect localization accuracy and completeness by employing optimized region growing operations to eliminate noise in defect regions. Moreover, we present an optimized inference method that provides meaningful visual explanations for defect recognition. Experimental results on real-world wafer map images demonstrate the effectiveness of our approach in accurately segmenting defect patterns with no pixel-level annotations. By training the model solely on wafer map image classification labels, our proposed model significantly improves defect recognition, facilitating efficient defect analysis in semiconductor manufacturing. The proposed weakly supervised learning approach offers a practical solution for defect diagnosis and localization, with the potential of widespread adoption in the semiconductor industry.

The entropy source structure with embedded XOR gates in a ring oscillator (RO) as a true random number generator (TRNG) can improve the speed of accumulating jitter in the oscillator. However, the XOR gate has a certain response time to the input change, and when the input changes too fast, the XOR gate will output short pulses. In this paper, we propose a TRNG design based on a multi-ring convergence oscillator (MRCO) making use of the characteristics of short pulses. We study the output of the XOR gate when facing different inputs. By modeling the time of a fibonacci ring oscillator (FIRO) as an example, we find that the loss of short pulses in an inverter chain is the reason for making the FIRO enter into periodic oscillation. This phenomenon suppresses the accumulation of jitter and occurs periodically in existing structures. Our proposed structure uses independent sub-rings to accumulate jitter, allowing the main-ring to quickly generate short pulses to provide analog randomness. The proposed TRNG design is implemented in Xilinx Virtex-6 FPGA. The experimental results show that it has the highest ratio of throughput rate to hardware resources. The generated random sequence pass both NIST SP800-22 test and NIST SP800-90B test.

Under the functional safety standard ISO26262, automotive systems require testing in the field, such as the power-on self-test (POST) . Unlike the production test, the POST requires reducing the test application time to meet the indispensable test quality (e.g., >90% of latent fault metric) of ISO26262. This article proposes a test point insertion technique for multi-cycle power-on self-test to reduce the test application time under the indispensable test quality. The main difference to the existing test point insertion techniques is to solve the fault masking problem and the fault detection degradation problem under the multi-cycle test. We also present the method to identify a user-specified amount of test points that could achieve the most scan-in pattern reduction for attaining a target test coverage. The experimental results on ISCAS89 and ITC99 benchmarks show 24.4X pattern reduction on average to achieve 90% stuck-at fault coverage confirming the effectiveness of the proposed method.

Multi-cycle test with partial observation for scan-based logic BIST is known as one of effective methods to improve fault coverage without increase of test time. In the method, the selection of flip-flops for partial observation is critical to achieve high fault coverage with small area overhead. This paper proposes a selection method under the limitation to a number of flip-flops. The method consists of structural analysis of CUT and logic simulation of test vectors, therefore, it provides an easy implementation and a good scalability. Experimental results on benchmark circuits show that the method obtains higher fault coverage with less area overhead than the original method. Also the relation between the number of selected flip-flops and fault coverage is investigated.

Multi-cycle Test applies more than one capture cycles during the capture operation which is a promising way to reduce the test volume of Logic-BIST (Logic Built-in Self-Test) based POST (Power-on Self-Test) for achieving high fault coverage. However, the randomness loss of the capture patterns due to the large number of capture cycles obstructs the further improvement of fault coverage and pattern reduction. In this paper, we propose a novel approach to control the capture patterns by modifying the captured values of scan Flip-Flops (FFs) during capture operation to enhance the test quality of the capture patterns. In the approach, we insert FF-Control circuits between the scan FFs and the combinational circuit to improve the randomness of the capture patterns by loading toggle vectors/pseudo-random vectors. The experimental results of ISCAS89 and ITC99 benchmarks validated the effectiveness of the proposed methods in fault coverage improvement and random pattern reduction for Logic-BIST.

To attain the requirement of ISO26262 standard, the POST for automotive MCU needs to achieve high Latent Fault (LF) metric (>90% for ASIL D) within limited test application time (TAT). In this paper, we propose a new DFT technique named Fault-Detection-Strengthened (FDS) method to enhance the effect of test pattern reduction of the multi-cycle test for shortening the TAT of POST, and develop an original in-house tool named FVP-TPI (Fault Vanishing Point-TPI) to implement the FDS method to automotive MCU. The evaluation results on a latest commercial automotive MCU (62M gates) confirm the effectiveness (test volume compaction) and the practicability (smaller hardware overhead, shorter period of DFT) of the method.

Recently, reconfigurable devices are gaining increased attention for the development of IoT, Automotive and AI system. A new type of fine-grained reconfigurable device named MRLD (Memory Based Reconfigurable Logic Device) has been proposed which is constructed by general SRAMs without any programmable interconnect resources. It should be a promising alternative to FPGA with the benefits of low production cost, low power and small delay. In this paper, we overview the architecture and the operation principle of MRLD. We also propose a test strategy and algorithms of pattern generation for the interconnect defects referred to stuck-at and bridge faults under MRLD. Experimental results confirmed the effectiveness of the proposed test method.

High power dissipation during scan-based logic BIST is a crucial problem that leads to over-testing. Although controlling test power of a circuit under test (CUT) to an appropriate level is strongly required, it is not easy to control test power in BIST. This paper proposes a novel power controlling method to control the toggle rate of the patterns to an arbitrary level by modifying pseudo random patterns generated by a TPG (Test Pattern Generator) of logic BIST. While many approaches have been proposed to control the toggle rate of the patterns, the proposed approach can provide higher fault coverage. Experimental results show that the proposed approach can control toggle rates to a predetermined target level and modified patterns can achieve high fault coverage without increasing test time.

BIST based field testing is a promising way to guarantee the functional safety of intelligent and autonomous systems. To improve the fault coverage with less random patterns for BIST, sequentially observing some flip-flops (FFs) during multi-cycle test is useful. In this paper, we propose the methodology for selecting the Fault-Detection-Strengthened FFs in multi-cycle test by evaluating the structure of a circuit. The experimental results of ITC99 benchmarks and a real Electronic Control Unit (ECU) circuit show the effectiveness of the proposed methods in fault coverage improvement and random pattern reduction.

Low-power test technology has been investigated deeply to achieve an accurate and efficient testing. Although many sophisticated methods are proposed for scan-test, there are not so many for logic BIST because of its uncontrollable randomness. However, logic BIST currently becomes vital for system debug or field test. This paper proposes a novel low power BIST technology that reduces shift-power by eliminating the specified high-frequency parts of vectors and also reduces capture power. The authors show that the proposed technology not only reduces test power but also keeps test coverage with little loss.

High test power in logic BIST is a serious problem not only for production test, but also for board test, system debug or field test. Many low power BIST approaches that focus on scan-shift power or capture power have been proposed. However, it is known that a half of scan-shift power is compensated by test responses, which is difficult to control in those approaches. This paper proposes a novel approach that directly reduces scan-out power by modifying some flip-flops' values in scan chains at the last capture. Experimental results show that the proposed method reduces scan-out power up to 30% with little loss of test coverage.