SVSoC: Speculative Vision Systems-on-a-Chip

Frame latency in continuous vision significantly impacts the agility of intelligent machines that interact with the environment via cameras. However, today's continuous vision systems limit the frame latency due to their fundamental sequential execution model. We propose a speculative execution model along with two mechanisms that enable practical vision speculation. We present SVSoC , a new mobile Systems-on-a-chip (SoC) architecture that augments conventional mobile SoCs with the speculation capability. Under the same energy budget, SVSoC achieves 14.3 to 35.4 percent latency reduction in different scenarios.