Approximate Communication for Energy Efficient Wireline Links

点击翻译按钮获取中文摘要

The goal of this project is to reduce the cost of a wireline communication system by improving its energy efficiency. In pursuing this goal, the project will investigate mechanisms to trade the accuracy of communication data bits with the energy efficiency of wireline links and analyze how to do this trade-off dynamically. The research is motivated by the fact that emerging data applications such as recognition, mining, search, data analytics, and machine learning are naturally resilient and can still produce correct output in the presence of erroneous data. The concept of approximate computing has leveraged this very resilient nature of data applications to improve the efficiency of computing. This project will leverage the same resilient nature to improve the energy efficiency of wireline data links by doing approximate communication. The results of this research will open a new dimension for improving energy efficiency by trading communication data-bit accuracy with energy efficiency. The research outcomes will give software developers a control knob with which energy-aware applications can be developed. The success of this research will motivate computer architects and compiler writers to provide energy-accuracy trade-off features to software developers with the aim of realizing an energy-efficient computing platform running energy-aware applications. This project will investigate control knobs and optimal parameters to achieve desired bit-error-rate (BER) with minimum energy per bit. Reconfigurable transceiver architectures investigated as part of this research will help support dynamic accuracy requirements of the data applications. This project will also investigate energy-scalable universal transmitter architectures, with the ability to support different pulse amplitude modulation (PAM) formats (PAM-4/8/16/32), different pulse width modulation (PWM) formats (PWM-4/8/16), and equalization methods on these modulations will be investigated. Furthermore, time-domain processing circuits at the receivers will be investigated, which can effectively trade-off energy consumption with accuracy. The approach of this project is unique because it focuses on improving energy efficiency through accuracy-energy trade-off, while other existing approaches focus on improving the efficiency of individual wireline circuits for a fixed accuracy. Energy-scalable techniques and architectures invented as part of this research can also be used to achieve approximate optical communication.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.