Significance driven hybrid 8T-6T SRAM for energy-efficient synaptic storage in artificial neural networks

Multilayered artificial neural networks have found widespread utility in classification and recognition applications. The scale and complexity of such networks together with the inadequacies of general purpose computing platforms have led to a significant interest in the development of efficient hardware implementations. In this work, we focus on designing energy-efficient on-chip storage for the synaptic weights, motivated primarily by the observation that the number of synapses is orders of magnitude larger than the number of neurons. Typical digital CMOS implementations of such large-scale networks are power hungry. In order to minimize the power consumption, the digital neurons could be operated reliably at scaled voltages by reducing the clock frequency. On the contrary, the on-chip synaptic storage designed using a conventional 6T SRAM is susceptible to bitcell failures at reduced voltages. However, the intrinsic error resiliency of neural networks to small synaptic weight perturbations enables us to scale the operating voltage of the 6T SRAM. Our analysis on a widely used digit recognition dataset indicates that the voltage can be scaled by 200 mV from the nominal operating voltage (950 mV) for practically no loss (less than 0.5%) in accuracy (22 nm predictive technology). Scaling beyond that causes substantial performance degradation owing to increased probability of failures in the MSBs of the synaptic weights. We, therefore propose a significance driven hybrid 8T-6T SRAM, wherein the sensitive MSBs are stored in 8T bitcells that are robust at scaled voltages due to decoupled read and write paths. In an effort to further minimize the area penalty, we present a synaptic-sensitivity driven hybrid memory architecture consisting of multiple 8T-6T SRAM banks. Our circuit to system-level simulation framework shows that the proposed synaptic-sensitivity driven architecture provides a 30.91% reduction in the memory access power with a 10.41% area overhead, for less than 1% loss in the classification accuracy.