Digital Perceptron Parallel Processing via Content-Addressable Memory
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional digital processors consume excessive energy and time due to sequential logic operations and frequent memory accessing, which is inefficient compared to the parallel processing and energy efficiency of biological neural networks.
Innovation Solution
A digital signal processor, named Digital Perceptron, uses a configurable non-volatile content memory array and CEEPROM array to process digital information in a parallel feed-forward manner, mimicking neural networks by storing and matching digital symbols directly in memory without sequential Boolean logic operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If sequential logic operations and frequent memory accessing are used in conventional digital processors, then computation can be performed according to pre-written instructions, but energy consumption and processing time increase significantly
Solution Approach 1:
The patent replaces the mechanical sequential execution mechanism of conventional CPUs with a content-addressable memory-based parallel processing system. Instead of fetching and executing instructions sequentially through arithmetic logic units, the system directly processes digital symbols by comparing input symbols with stored symbols in content-addressable memory, eliminating the need for sequential instruction fetching, decoding, and execution cycles.
Solution Approach 2:
The patent pre-configures digital symbols and their corresponding output symbols in non-volatile content-addressable memory before processing begins. This preliminary configuration allows the system to perform parallel pattern matching without sequential computation, as all possible input-output relationships are pre-established in the memory structure, enabling direct lookup and processing.
2Ease of operation
If sequential instruction execution is used in Von Neumann architecture, then logic operations can be performed, but the frequency of memory accessing increases, consuming more energy and time
Solution Approach 1:
The patent merges the instruction storage, data storage, and processing functions into a single content-addressable memory structure. Instead of separate instruction memory, data memory, and arithmetic logic units that require multiple access cycles, the system combines these functions so that processing occurs directly within the memory structure through parallel symbol comparison and matching.
3Adaptability or versatility
If conventional digital processors use sequential Boolean logic operations, then computations can be performed according to programmed instructions, but the processing efficiency is inferior to biological neural networks
Solution Approach 1:
The patent changes the fundamental processing parameter from sequential Boolean logic operations to parallel content-based symbol matching. Instead of using Boolean algebra and sequential arithmetic operations, the system uses content-addressable memory to perform parallel comparison of digital symbols with stored patterns, fundamentally changing how computation is performed to achieve neural network-like efficiency.
Data Source
AI summary
In view of the neural network information parallel processing, a digital perceptron device analogous to the build-in neural network hardware systems for parallel processing digital signals directly by the processor's memory content and memory perception in one feed-forward step is disclosed. The digital perceptron device of the invention applies the configurable content and perceptive non-volatile memory arrays as the memory processor hardware. The input digital signals are then broadcasted into the non-volatile content memory array for a match to output the digital signals from the perceptive non-volatile memory array as the content-perceptive digital perceptron device.