Predictive Window Decoding to Minimize Program Runtime

Real-time decoding is a key ingredient in future fault-tolerant quantum systems. Prior work has shown parallel window schemes can scalably meet throughput requirements in the presence of increasing decoding times given enough classical resources. However, such schemes introduce sub-optimal decoding latency, resulting in additional delay during T injection that slows down program performance. To alleviate this, we introduce predictive window decoding. Our scheme utilizes a light-weight prediction step that allows windows to begin decoding soon after they are generated. As a result, the decoding latency, and critically the latency of blocking operations such as T gates, is reduced, leading to a shorter program runtime. In our evaluation, we compare window decoding with and without prediction for a range of fault-tolerant benchmark applications, and we analyze the sensitivity of our results to prediction accuracy.