This paper presents a novel architecture for in-memory computation of binary neural network (BNN) workloads based on STT-MRAM arrays. In the proposed architecture, BNN inputs are fed through bitlines, then, a BNN vector multiplication can be done by single sensing of the merged SL voltage of a row. Our design allows to perform unrestricted accumulation across rows for full utilization of the array and BNN model scalability, and overcomes challenges on the sensing circuit due to the limitation of low regular tunneling magnetoresistance ratio (TMR) in STT-MRAM. Circuit techniques are introduced in the periphery to make the energy-speed-area-robustness tradeoff more favorable. In particular, time-based sensing (TBS) and boosting are introduced to enhance the accuracy of the BNN computations. System simulations show 80.01% (98.42%) accuracy under the CIFAR-10 (MNIST) dataset under the effect of local and global process variations, corresponding to an 8.59% (0.38%) accuracy loss compared to the original BNN software implementation, while achieving an energy efficiency of 311 TOPS/W.

In this paper, a row-wise XNOR accumulator architecture for STT-MRAM arrays is proposed for parallel and efficient multiply-and-accumulate (MAC) operation. The proposed accumulator supports in-memory computing and binary neural network (BNN) applications. In the proposed architecture, inputs are fed from the complementary bitlines, whereas readout is performed through a time-based sense amplifier (TBS). The proposed architecture that does not require any ADC can exhibit an average error rate of 0.085 for XNOR vector size (i.e., accumulate capacity) of 128 bits, which translates into 98.45% classification accuracy of a multi-layer perceptron (MLP) on the MNIST dataset.

No abstract

Binary neural networks (BNNs) that use 1-bit weights and activations have garnered interest as extreme quantization provides low power dissipation. By implementing BNNs as computing-in-memory (CIM), which computes multiplication and accumulations on memory arrays in an analog fashion, namely analog CIM, we can further improve the energy efficiency to process neural networks. However, analog CIMs suffer from the potential problem that process variation degrades the accuracy of BNNs. Our Monte-Carlo simulations show that in an SRAMbased analog CIM of VGG-9, the classification accuracy of CIFAR-10 is degraded even below 20% under process variations of 65nm CMOS. To overcome this problem, we present a variation-aware BNN framework. The proposed framework is developed for SRAM-based BNN CIMs since SRAM is most widely used as on-chip memory, however easily extensible to BNN CIMs based on other memories. Our extensive experimental results show that under process variation of 65nm CMOS, our framework significantly improves the CIFAR-10 accuracies of SRAM-based BNN CIMs, from 10% and 10.1% to 87.76% and 77.74% for VGG-9 and RESNET-18 respectively.