This article presents a 6-bit 1.1 GS/s asynchronous charge-injection SAR (ci-SAR) analog-to-digital-converter (ADC) achieving an unprecedented area efficiency of 0.0048 GS/s/ $\mu$ m $^{2}$ per slice. A current-reduced charge injection cell (CIC) enables a small input capacitance of 8.5 fF, which helps extend the effective resolution bandwidth (ERBW) to 7 GHz. A replica-based feedback loop desensitizes the full-scale variations on process, voltage, and temperature (PVT) conditions. A common-mode shifting scheme relaxes the dynamic offset issue of the comparator inherent in monotonic switching. The chip is fabricated in a 28 nm CMOS process with an active area of 227 $\mu$ m $^{2}$ . The peak SNDR is 35.2 dB with FOM $_{\text{w}}$$=$ 25.5 fJ/conv.-step. The measured SNDR drops by less than 2.32 dB over $-$ 40 $^{\circ}$ C to 100 $^{\circ}$ C temperature variation and 0.9 to 1.1 V supply variation.
14 June 2023 in IEEE Transactions on Circuits and Systems II: Express Briefs
This paper presents a 32x32 pseudo-ReRAM-based analog computing-in-memory (CIM) macro in 28nm CMOS. A 4b self-error-correcting word-line (WL) driver reduces the analog compute inaccuracy while minimizing the latency. A stability compensating dummy row maximizes the accumulation length of the multiply-and-accumulate (MAC). The columnsensing dual-phase 6b successive-approximation-register (SAR) analog-to-digital-converter (ADC) maximizes the through-put with minimized pitch. The proposed CIM occupies an active area of 0.0155mm2 and consumes 4.36mW with an average energy efficiency of 25.8TOPS/W. The measured performance achieves the highest normalized throughput with an end-to-end inference accuracy comparable to FP32 with less than a 0.11% drop.
An 8-bit switched-capacitor multiply-and-accumulator (MAC) in 65nm CMOS is presented. Based on a cascaded low-power ring-amplifier-based switched-capacitor DACs, the MAC circuit features a programmable accumulation length in MAC computation. Fabricated in 65nm CMOS, the prototype MAC circuit achieves a precision-scaled energy efficiency of 1.32fJ per MAC operation, which is comparable to other state-of-the-art MAC circuits, along with best-in-class linearity. The noise performance has been verified using four real-world convolutional neural networks (CNNs) and 10,000-image data sets with up to 1,000 classes with an accuracy drop of less than 2% compared to the baseline 32-bit floating-point MAC.