Realizing increasingly complex artificial intelligence (AI) functionalities directly on edge devices calls for unprecedented energy efficiency of edge hardware. Compute-in-memory (CIM) based on resistive random-access memory (RRAM)1 promises to meet such demand by storing AI model weights in dense, analogue and non-volatile RRAM devices, and by performing AI computation directly within RRAM, thus eliminating power-hungry data movement between separate compute and memory2–5. Although recent studies have demonstrated in-memory matrix-vector multiplication on fully integrated RRAM-CIM hardware6–17, it remains a goal for a RRAM-CIM chip to simultaneously deliver high energy efficiency, versatility to support diverse models and software-comparable accuracy. Although efficiency, versatility and accuracy are all indispensable for broad adoption of the technology, the inter-related trade-offs among them cannot be addressed by isolated improvements on any single abstraction level of the design. Here, by co-optimizing across all hierarchies of the design from algorithms and architecture to circuits and devices, we present NeuRRAM—a RRAM-based CIM chip that simultaneously delivers versatility in reconfiguring CIM cores for diverse model architectures, energy efficiency that is two-times better than previous state-of-the-art RRAM-CIM chips across various computational bit-precisions, and inference accuracy comparable to software models quantized to four-bit weights across various AI tasks, including accuracy of 99.0 percent on MNIST18 and 85.7 percent on CIFAR-1019 image classification, 84.7-percent accuracy on Google speech command recognition20, and a 70-percent reduction in image-reconstruction error on a Bayesian image-recovery task.

No abstract

No abstract

Implantable and ambulatory measurement of physiological signals such as Bio-impedance using miniature biomedical devices needs careful tradeoff between limited power budget, measurement accuracy and complexity of implementation. This paper addresses this tradeoff through an extensive analysis of different stimulation and demodulation techniques for accurate Bio-impedance measurement. Three cases are considered for rigorous analysis of a generic impedance model, with multiple poles, which is stimulated using a square/sinusoidal current and demodulated using square/sinusoidal clock. For each case, the error in determining pole parameters (resistance and capacitance) is derived and compared. An error correction algorithm is proposed for square wave demodulation which reduces the peak estimation error from 9.3% to 1.3% for a simple tissue model. Simulation results in Matlab using ideal RC values show an average accuracy of for single pole and for two pole RC networks. Measurements using ideal components for a single pole model gives an overall and readings from saline phantom solution (primarily resistive) gives an . A Figure of Merit is derived based on ability to accurately resolve multiple poles in unknown impedance with minimal measurement points per decade, for given frequency range and supply current budget. This analysis is used to arrive at an optimal tradeoff between accuracy and power. Results indicate that the algorithm is generic and can be used for any application that involves resolving poles of an unknown impedance. It can be implemented as a post-processing technique for error correction or even incorporated into wearable signal monitoring ICs.