Digital-analog mixed signal processing system for imprecise computation

Abstract

The invention discloses a digital-analog mixed signal processing system for imprecise computation, comprising an input module used for decomposing A B-bit digital input signals into A*B single-bit high- and low-level signals and sending the high- and low-level signals to a signal processing module as analog input signals, the signal processing module used for calculating the analog input signals to obtain C analog output signals and sending the analog output signals to an output module, and the output module used for converting the analog output signals into D high- and low-level signals and outputting the D high- and low-level signals as E F-bit digital signals, wherein D, E and F are natural numbers, and D=E*F. The digital-analog mixed signal processing system for imprecise computation has the advantages as follows: the area and power consumption of a computing system on a conversion interface are reduced effectively, and interface design is greatly simplified; and the area is reduced by more than 50%, the energy consumption is reduced by more than 60%, and the application space of the system of the type is significantly improved.

Description

technical field [0001] The invention relates to the technical field of computers and electronic information, in particular to a digital-analog mixed signal processing system for brain-like calculations and approximate calculations. Background technique [0002] With the continuous development of the information age, people's demand for computational complexity and data scale has exploded, and energy consumption has become an important bottleneck that restricts the performance of computing systems. However, as the development of traditional CMOS technology has approached its physical limit, it has become more and more difficult to improve the energy efficiency of the system from the process of the contemporary digital system. At the same time, in the traditional Von Neumann architecture, the storage bandwidth required by high-performance CPUs has gradually exceeded the effective bandwidth that traditional storage architectures can provide, thus creating a new communication bo...

AI Technical Summary

Problems solved by technology

However, compared with analog computing circuits such as memristor arrays with high integration and high energy efficiency, these digital-analog / analog-digital conversion circuits not only occupy a large area, but also consume far more energy than analog circuits.

Experimental data shows that the occupied area and energy consumption of the digital-analog / analog-digital conversion circuit can reach more than 85% of the entire system, which makes the digital-analog mixed signal processing system based on non-accurate calculation in terms of energy consumption and area Advantages and potentials are wasted which in turn limits the further development of such systems

Images