Neuromorphic Arithmetic Device Offset Error Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neuromorphic arithmetic devices face reliability and recognition rate degradation due to offset errors during extensive computation, which existing technologies fail to effectively address.
Innovation Solution
A neuromorphic arithmetic device with an offset accumulator, bit extractor, and cumulative synapse array that measures and corrects offset data by accumulating and averaging offset values, allowing for accurate MAC calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If analog MAC operation is used for neural processing, then low power and low area are achieved, but offset errors degrade reliability and recognition rate
Solution Approach 1:
The patent applies preliminary action by measuring offset errors before actual neural network computation and storing these offset values in an offset register. This pre-measurement and pre-storage of offset data allows the system to compensate for errors during MAC operations without adding real-time correction complexity, thus maintaining low power consumption while improving reliability and recognition rate through advance error characterization
Solution Approach 2:
The patent implements feedback by using measured offset errors to correct subsequent MAC operation results. The offset values obtained through preliminary measurement are fed back into the computation process to compensate for systematic errors, creating a closed-loop correction mechanism that improves reliability without significantly increasing power consumption
2Reliability
If offset correction is implemented in neuromorphic arithmetic devices, then reliability and recognition rate are improved, but device complexity increases
Solution Approach 1:
The patent extracts offset errors as a separate, measurable quantity from the main computation process. By isolating the offset measurement function into distinct measurement operations and storing offset values in a dedicated register, the system simplifies the correction mechanism while improving reliability, avoiding the need for complex real-time correction circuits
Solution Approach 2:
The patent introduces an intermediary offset register that stores pre-measured offset values. This intermediary structure mediates between the measurement process and the MAC computation, allowing offset correction without direct complex interaction between measurement and computation units, thus reducing overall device complexity while maintaining improved reliability
Data Source
AI summary
The neuromorphic arithmetic device performs a multiply-accumulate (MAC) calculation using a multiplier and an accumulator. The neuromorphic arithmetic device includes an offset accumulator configured to receive a plurality of offset data measured a plurality of times and accumulate the plurality of offset data, a bit extractor configured to obtain average offset data by extracting at least one first bit from the plurality of accumulated offset data, and a cumulative synapse array configured to accumulate a plurality of multiplication values generated by the multiplier and output a cumulative result of the plurality of multiplication values corrected according to the average offset data.