19 November 2024 in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
Deep neural network (DNN) accelerators are increasingly integrated into sensing applications, such as wearables and sensor networks, to provide advanced in-sensor processing capabilities. Given wearables’ strict size and power requirements, minimizing the area and energy consumption of DNN accelerators is a critical concern. In that regard, computing DNN models in the time domain is a promising architecture, taking advantage of both technology scaling friendliness and efficiency. Yet, time-domain accelerators are typically not fully digital, limiting the full benefits of time-domain computation. In this work, we propose an all-digital time-domain accelerator with a small size and low energy consumption to target precision in-sensor processing like human activity recognition (HAR). The proposed accelerator features a simple and efficient architecture without dependencies on analog nonidealities such as leakage and charge errors. An eight-neuron layer (core computation layer) is implemented in 22-nm FD-SOI technology. The layer occupies $70 \times\,70\,\mu$ m while supporting multibit inputs (8-bit) and weights (8-bit) with signed accumulation up to 18 bits. The power dissipation of the computation layer is 576 $\mu$ W at 0.72-V supply and 500-MHz clock frequency achieving an average area efficiency of 24.74 GOPS/mm $^2$ (up to 544.22 GOPS/mm $^2$ ), an average energy efficiency of 0.21 TOPS/W (up to 4.63 TOPS/W), and a normalized energy efficiency of 13.46 1b-TOPS/W (up to 296.30 1b-TOPS/W).
26 May 2025 in IEEE Internet of Things Journal
As the global population ages, caregivers encounter significant challenges in monitoring wet diapers and assessing the volumes of voided fluids in adult incontinence products. Recent advancements in smart-diaper technology address these issues, but challenges persist with integrating electronics, frequent removal and reapplication, and managing batteries. This study presents a batteryless urine-powered IoT sensor node using printed energy harvesters (EH) and capacitive sensors integrated with an ultralow-power integrated circuit for disposable smart diapers. Coplanar capacitive sensors and electrochemical EH are printed on the flexible substrate using environment-friendly materials. An ultralow power frontend interface is developed, which is powered by the harvested energy from urine. Laboratory measurements validate the functionality of on-chip electronics and the performance of in-diaper sensors and EH. The system demonstrates effective energy harvesting by maintaining a stable regulated voltage of 1.1 V with urine volumes of 90 ml or more, powering the front-end electronics continuously for 6 h. The proposed system successfully demonstrated the detection of multiple urination events in the diaper and quantified the voided volume as low as 30 ml. The results demonstrated in this work pave the way for cost-effective, disposable, and environmentally friendly solutions for smart diapers, enhancing both efficiency and comfort for caregivers and elderly individuals.
8 July 2025 in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
in-memory computing (IMC) has emerged as one of the most promising architectures to efficiently compute artificial intelligence tasks on hardware, particularly deep neural networks (DNNs). IMC can make use of analog computation principles alongside emerging nonvolatile memories (eNVM) technologies, potentially offering several orders of magnitude increased energy efficiency compared to generic processing units. Yet, the use of analog circuitry, potentially integrated with emerging technologies post-processed on top of silicon wafers, increases the susceptibility of hardware to a large spectrum of variations, for instance manufacturing, noise or temperature sensitivity. Hence, this susceptibility can hamper the large-scale deployment of IMC circuits into the market. To tackle the reliability of analog resistive-based IMC circuits regarding temperature variations, this article presents TRIM, a thermal on-chip auto-compensation method aimed at fully calibrating first-order temperature effects. TRIM is designed to maintain the computational accuracy of IMC cores in DNN applications over a wide temperature range, while being highly scalable and adaptable. In essence, the temperature compensation is realized through a complementary-to-absolute-temperature (CTAT) voltage reference integrated inside a voltage regulator and applied at the zero reference node of a multiplying digital-to-analog converter (MDAC), eliminating the need for external circuits or look-up table. The proposed methodology is demonstrated on a proof-of-concept 65 nm CMOS resistive IMC column. Measurement results showcase that the proof-of-concept auto-compensation system significantly enhances inference and multiply-and-accumulate (MAC) operation accuracy of any first-order resistive crossbar column, achieving inference accuracy recovery of 100% over a temperature range of –20 °C to 60 °C and a 91.3% improvement in MAC operation accuracy, with an area overhead of 2% and power overhead of ¡ 0.02%.
1 November 2026 in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
The development of accurate and robust compute-in-memory (CIM) architectures is a central research focus for accelerating artificial intelligence (AI) tasks, particularly deep neural networks (DNNs). Significant interest has emerged in analog and mixed-signal CIM architectures to improve the efficiency of data storage and computation, as well as to manage the large data volumes required by DNNs. Recent advances in emerging non-volatile memory (eNVM) solutions have driven progress in resistive mixed-signal CIM cores. However, mixed-signal CIM computing cores continue to face integration and robustness challenges that limit their widespread adoption in end-to-end AI computing systems. From an integration perspective, resistive and eNVM-based CIM cores require integration with a control processor to enable programmable acceleration. Additionally, SRAM-based CIM architectures remain more efficient and easier to program than eNVM-based alternatives. In terms of robustness, analog circuits are more susceptible to variations, leading to computation errors and reduced accuracy. This study addresses both challenges by introducing a self-calibrated mixed-signal CIM accelerator system-on-chip (Acore-CIM), fabricated using 22-nm FD-SOI technology. Integration is achieved through (1) a CIM architecture that combines the density and programmability of SRAM-based weight storage with multi-bit computation using linear resistors, alongside (2) an open-source programming and testing strategy for CIM systems. Accuracy and robustness are enhanced through automated RISC-V-controlled on-chip calibration, which improves compute signal-to-noise ratio (SNR) by 25–45 % across columns (reaching 18–24 dB) and reduces process-induced multiply-and-accumulate (MAC) error spread by up to 82 %. To demonstrate scalability, the proof-of-concept system-on-chip (SoC) is shown to be extendable to recent high-density linear resistor technologies, further enhancing computing performance.