Technology Category
Machine Learning: Hardware-aware MLCognitive Modeling & Cognitive Systems: (Computational) Cognitive ArchitecturesComputer Vision: Adversarial Attacks & Robustness
4T2R X-ReRAM CiM Array for Variation-tolerant, Low-power, Massively Parallel MAC Operation
🤖 AI Summary
To address the challenges of escalating power consumption and computation errors induced by device variations in high-row-parallel compute-in-memory (CiM) architectures, this work proposes a low-power, robust multiply-accumulate (MAC) architecture for AI accelerators. We introduce a novel 4T2R ReRAM-based CiM cell—reducing device count by 50% versus conventional 4T4R designs—thereby mitigating process variation impact and lowering write energy. Complementing this, an optimized 8T SRAM-assisted analog-domain accumulation circuit enables precise, high-parallelism summation. Experimental results demonstrate that, while maintaining a 128×128 MAC parallelism scale, the proposed architecture achieves a 2.3× improvement in energy efficiency and reduces computational error rate to 0.8%, significantly enhancing system fault tolerance and operational stability. This work establishes a scalable hardware paradigm for high-reliability CiM accelerators.
Technology Category
Application Category
📝 Abstract
Computation-in-Memory (CiM) is attracting attention as a technology that can perform MAC calculations required for AI accelerators, at high speed with low power consumption. However, there is a problem regarding power consumption and device-derived errors that increase as row parallelism increases. In this paper, a 4T2R ReRAM cell and an 8T SRAM CiM suitable for CiM is proposed. It is shown that adopting the proposed 4T2R ReRAM cell reduces the errors due to variation in ReRAM devices compared to conventional 4T4R ReRAM cells.
Problem
Research questions and friction points this paper is trying to address.
Reduces power consumption in Computation-in-Memory arrays
Mitigates device-derived errors in ReRAM-based MAC operations
Improves variation tolerance compared to conventional 4T4R cells
Innovation
Methods, ideas, or system contributions that make the work stand out.
4T2R ReRAM cell for variation tolerance
8T SRAM CiM for low power
Massively parallel MAC operation
🔎 Similar Papers
No similar papers found.
