PREDITOR: Enhancing Error Resilience of a Near-Threshold Tensor Processing Unit
TimeTuesday, December 7th6:00pm - 7:00pm PST
Event Type
Networking Reception
Work-in-Progress Poster
Virtual Programs
Presented In-Person
DescriptionIncreasing processing requirements in the Artificial Intelligence realm has led to the emergence of Domain specific architectures for Deep Neural Network (DNN) applications. Tensor Processing Unit (TPU), a DNN accelerator by Google has emerged as a front-runner outclassing its contemporaries, CPUs and GPUs in performance by 15x-30x. However, a TPU’s performance enhancement is accompanied by a mammoth power consumption. In the pursuit of lowering the energy utilization, we propose PREDITOR—a low power TPU, operating in the Near-Threshold Computing realm. Preditor selectively boosts the voltage of selective multiplier-and-accumulate units, offering upto 3x-5x improved performance with a minor loss in accuracy.