Programmable Offset Cells for Compute-in-Memory Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Compute-in-memory (CIM) architectures face challenges in mitigating noise and inaccuracies introduced by conversions between digital-domain and analog-domain components, particularly in resource-limited devices like mobile and IoT devices, which are used in neural network inferencing applications.
Innovation Solution
The implementation of a CIM array with programmable offset cells and an analog-to-digital converter (ADC) that receives a programmable offset value, allowing for improved accuracy and reduced power consumption by minimizing noise during conversions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If conversions between digital-domain and analog-domain components are performed in CIM architecture, then computational power is improved, but noise and inaccuracies are introduced
Solution Approach 1:
The patent applies preliminary action by performing offset calibration before actual neural network computations. The system pre-determines offset values for each column of the CIM array that compensate for systematic errors in the analog-domain computations. This preliminary calibration step ensures that subsequent conversions between digital and analog domains produce more accurate results, reducing the noise and inaccuracies introduced by these conversions while maintaining the computational power benefits of CIM architecture.
2Use of energy by moving object
If resource-limited devices use CIM architecture for neural network inferencing, then power consumption is reduced, but accuracy deteriorates due to conversion noise
Solution Approach 1:
The patent applies parameter changes by dynamically adjusting the offset values for each column of the CIM array based on calibration measurements. The system modifies the reference parameters (offset values) to compensate for variations in analog-domain computations. This allows resource-limited devices to maintain higher inferencing accuracy while utilizing the low power consumption benefits of CIM architecture, as the adjusted offset parameters correct the accuracy deterioration caused by conversion noise.
Data Source
AI summary
In one embodiment, an electronic device includes a compute-in-memory (CIM) array that includes a plurality of columns. Each column includes a plurality of CIM cells connected to a corresponding read bitline, a plurality of offset cells configured to provide a programmable offset value for the column, and an analog-to-digital converter (ADC) having the corresponding bitline as a first input and configured to receive the programmable offset value. Each CIM cell is configured to store a corresponding weight.