Resistive random-access memory (ReRAM) offers a potential solution to accelerate the inference of deep neural networks by performing processing-in-memory. However, the peripheral circuits of ReRAM crossbars used to perform arithmetic operations consume significant amounts of power. Based on a power consumption analysis of the ReRAM crossbar circuits, we propose using the dynamic reference voltage scalable analog-to-digital circuits (ADCs) to conduct the dot product operation to enable the reconfigurability of the ReRAM-based neural network (NN) accelerator while maintaining accuracy. We propose a configurable ReRAM-based NN accelerator to provide various degrees of computing granularity with different levels of power consumption, creating a tradeoff between performance and power consumption in the given NN. Next, we develop an energy-efficient inference engine for the configurable ReRAM-based NN accelerator, EIF, to assign the operation unit (OU) size to perform vector-matrix multiplication (VMM) based on the data dependence of the NN. Our evaluation shows that the proposed EIF provided an energy savings of up to 36% over the state-of-the-art ReRAM-based accelerator while maintaining performance without resource duplication.

Resistive random-access memory (ReRAM) has been explored to be a promising solution to accelerate the inference of deep neural networks at the embedded systems by performing computations in memory. To reduce the latency of the neural network, all the pre-trained weights are pre-programmed in ReRAM cells as device resistance for the inference phase. However, the system utilization is decreased by the data dependency of the deployed neural networks and results in low energy efficiency. In this work, we propose a Lazy Engine for providing high utilization and energy-efficient ReRAM-based accelerators. Instead of avoiding idle time by applying ReRAM crossbar duplication, Lazy Engine delays the start time of the vector-matrix multiplication operations, with run-time programming overhead consideration, to reclaim idle time for energy efficiency while improving resource utilization. The experimental results show that Lazy Engine achieves up to 77% and 96% improvement in resource utilization and energy saving compared to state-of-the-art ReRAM-based accelerators.