Presentation
TRAM: Training Approximate Multiplier Structures for Low-Power AI Accelerators
DescriptionReducing power consumption in AI accelerators is increasingly important. Approximate computing can reduce power consumption while keeping the accuracy loss small. Since multipliers are power-consuming in AI models, this paper focuses on synthesizing low-power approximate multipliers (AxMs). Unlike prior works that rely on manual or automatic synthesis without considering AI model contexts, we present TRAM, which jointly optimizes the AxM structure and AI model to lower power with small accuracy loss. Experiments show that compared to state-of-the-art AxMs, TRAM achieves up to 25.05% AxM power reduction on CNNs with CIFAR-10, and reduces power by 27.09% on vision transformers with ImageNet.
Event Type
Research Manuscript
TimeWednesday, July 2912:03pm - 12:16pm PDT
LocationMtg Room 201B