Approximate Communication
Recent research has shown that on-chip data movement consumes much more power than computation and this trend will continue in the future. Additionally, some algorithms and applications, such as machine learning, big data analysis, image/video processing and computer vision are tolerant to modest errors.
These applications are inherently tolerant to some error in their output whether due to noisy input data, multiple correct answers or not requiring very accurate execution. Leveraging this error tolerance can lead to significant energy savings and performance improvements.
As error correction techniques in data transmission, such as error correction codes and retransmission, consume energy and increase network latency, reducing the amount of error correction can significantly improve network performance. This project exploits the fact that not all applications require strong error correction and investigates approximate communication with dynamic error correction methods to trade-off absolute accurate data transmission for the power savings and transmission time.