Resilient circuits based on in situ timing monitoring adaptive voltage–frequency scaling (AVFS) eliminate excess time margins caused by process, voltage, and temperature (PVT) variations but suffer from 50% throughput loss during error recovery when operating at a half frequency. We propose a bi-directional adaptive clocking circuit to provide fine frequency tuning with zero latency for AVFS system. It can either stretch the clock cycle when there are timing errors to ensure correct function or compress the cycle when there are excess timing margins. To support a wide frequency range, we generate multiple phase clocks based on two delay lines and select one appropriate phase clock to obtain a stretched output clock, where balanced clock paths are obtained by a time-to-digital converter and dynamic-OR gates. Applied to a wide-operating-range AVFS system of an SHA-256 accelerator with transition detector (TD) latches, the whole AVFS system is able to respond to errors in one clock cycle, with dynamic-OR gates collecting all the errors in half a cycle and adaptive clocking circuit stretching at the current cycle. Fabricated in 28-nm CMOS, chip measurement shows that it achieves 38.6%–69.4% power gains at near threshold while reducing throughput loss during error recovery.
26 July 2023 in IEEE Transactions on Circuits and Systems II: Express Briefs
With transistors scaling down, aging effects become increasingly significant in circuit design. Thus, the aging-aware cell timing model is necessary for evaluating the aging-induced delay degradation and their impact on circuit performance. However, the tradeoff between accuracy and efficiency becomes a bottleneck in traditional methods. In this brief, we propose a fast and accurate aging-aware cell timing model via graph learning. The information of multi-typed devices on different arcs can be embedded by heterogeneous graph attention networks (HGAT) and the embedded results help improve the accuracy of our aging-aware timing model. The experimental results indicate the proposed timing model can achieve high accuracy efficiently.
22 December 2020 in IEEE Transactions on Circuits and Systems I: Regular Papers
Because of the different working principles between conventional flyback and active-clamped flyback (ACF), the existing primary-side regulation (PSR) technology cannot be applied to ACF. And in order to achieve PSR of ACF, a low-cost digital sampling method based on the auxiliary winding of the transformer is proposed. In the conventional peak current mode (PCM) controlled ACF, the sampling resistor of primary current may introduce additional power loss, and high-frequency oscillations during sampling may reduce system’s stability. Therefore, a digital feedback-feedforward control (FFC) method is proposed to eliminate the sampling resistor. The stability of the proposed control method and its influence on the dynamic response are studied. In addition, since GaN device has higher power loss during reverse conduction than its Si counterpart, design considerations of reverse conduction time of power switches are also discussed in this paper. All control methods and design considerations are verified on a GaN-based 12V-3A ACF, and the proposed control strategy is implemented by FPGA.
17 May 2023 in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
In advanced technology nodes, aging effects like negative and positive bias temperature instability (NBTI and PBTI) become increasingly significant, making timing closure and optimization more challenging. Unfortunately, conventional critical path (CP) selection tools used in reliability-aware design flow cannot accurately identify CPs under different aging conditions. To address this issue, we propose an aging-aware CP selection flow comprising two parts: 1) critical cell detection and 2) path criticality (PC) computation. We employ graph-attention (GAT) networks to predict the critical cells in the aged circuits, and a PC computation algorithm that takes into account circuit-level and transistor-level parameters to generate PC rank lists. Our experimental results demonstrate that our GAT model outperforms classical machine learning models in detecting critical cells. Additionally, compared with the commercial tool, our aging-aware flow achieves an average accuracy of 99.52%, 98.69%, and 97.20% for top-10%, top-5%, and top-1% path sets, respectively, in five industrial designs subjected to different aging conditions and workloads.
22 June 2022 in IEEE Journal of Emerging and Selected Topics in Power Electronics
In an active clamp flyback (ACF) converter working in critical mode (CRM), the zero-voltage switching (ZVS) of the main power switch can be realized by the negative valley current, which is related to conduction loss and switching loss. Therefore, how to control the negative valley current will affect the overall efficiency of the system. This article investigates and compares the three existing technologies and summarizes their limitations in practical applications. Therefore, a digital adaptive control method is proposed to regulate the valley current according to different working conditions. Meanwhile, small-signal analysis is performed to ensure the stability of the proposed control method. The proposed control method is verified by a field-programmable gate array (FPGA)-based 12-V, 3-A ACF, and experimental results show that the proposed theoretical model should be accurate and practical and there is an average efficiency improvement of 1.34% compared to conventional control.
This paper proposes an Energy-Efficient Reconfigurable Architecture (E-ERA) for Recurrent Neural Networks (RNNs). In E-ERA, reconfigurable computing arrays with approximate multipliers and dynamically adaptive accuracy controlling mechanism are implemented to achieve high energy efficiency. The E-ERA prototype is implemented on TSMC 45 nm process. Experimental results show that, comparing with traditional designs, the power consumption of E-ERA is reduced by 28.6%∼52.3%, with only 5.3%∼9.2% loss in accuracy. Compared with state-of-the-art architectures, E-ERA outperforms up to 1.78X in power efficiency and can achieve 304 GOPS/W when processing RNNs for speech recognition.