12 January 2022 in Nature
20 November 2023 in IEEE Journal of Solid-State Circuits
This article presents multistep accumulation capacitor coupling static random-access memory (MACC-SRAM), capacitor-based in-memory computing (IMC) SRAM macro for 4-b deep convolutional neural network (DNN) inference. The macro can simultaneously activate all its 128 $\times$ 128 custom 9T1C bitcells to perform the vector–matrix multiplication (VMM). MACC-SRAM also integrates 128 stepwise-charging and discharging input drivers (SCD-IDRs) to efficiently convert the digital codes of the input activations into analog voltages in a 2-b serial fashion. As a result, it can save up to 66% of the capacitor-driving energy. Also, the macro adopts an adder-first architecture to reduce the analog-to-digital (A/D) conversion overhead for the analog-mixed-signal (AMS) computation. The partial sums of the four adjacent rows, representing different bit positions in the 4-b weights, are first accumulated with an analog switched-capacitor adder and then converted to digital codes by a 6-bit successive approximation register (SAR) analog-to-digital converter (ADC). Compared with the ADC-first architecture, where partial sums of each row are first converted to digital codes and then accumulated in the digital domain, the adder-first architecture can save 60.7% of the area and 66.7% of the energy consumption of the A/D conversion. Moreover, the co-optimization of the DNN model by increasing the sparsity further reduces 39.4% of the capacitor-driving energy with a neglectable 0.4% DNN accuracy loss. We prototyped the macro in 28 nm technology, and the measurement shows an energy efficiency of 163 tera-operations per second per watt (TOPS/W) and a throughput of 211 giga-operations per second (GOPS) for a 4-b/4-b DNN model under 0.9 V supply, among the highest of the recent IMC SRAMs.
23 September 2024 in IEEE Solid-State Circuits Letters
This work presents an analog in-memory computing (IMC) macro utilizing 14nm MRAM technology. To facilitate energy-efficient high-throughput multiply-accumulate (MAC) operations, a multi-bit weight is introduced using stacked MTJ architecture and an analog bit-parallel MAC (ABP-MAC) scheme is proposed. This approach delivers 3.3 times better TOPS/mm2 than the state-of-the-art MRAM-based IMC macro. Additionally, a comprehensive calibration technique significantly improves computational accuracy across 72 IMC macros. The proposed IMC macro achieves 18.29 TOPS/mm2 and 340.8 TOPS/W with 1-bit normalization and classification accuracy of 90.2% with the Google Speech Commands dataset.
22 January 2025 in IEEE Solid-State Circuits Letters
Two specialized digital SRAM in-memory computing (IMC) macros were implemented using a 5nm process: (1) a high-efficiency (HE) macro, and (2) a high-density (HD) macro. The HE macro achieves an energy efficiency of 274 TOPS/W under 90% input bit sparsity, 50% weight bit sparsity, and 0.46V supply voltage, by adopting a compact Wallace tree adder and a tri-state buffer optimized for pipelined operation. The HD macro achieves a state-of-the-art memory density performance of 5.67 Mb/mm2 by employing a multiply-cell consisting of a single transistor. Extensive sample measurements have confirmed the robust and reliable performance of the two digital IMC macros. The proposed HE and HD macro have demonstrated their significant potential as key building blocks for next-generation AI accelerators.