In-Memory Computing Macro for SNN/ANN Mode Switching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computing architectures face limitations in performance and power efficiency due to frequent data movements between processor and memory, which hinder the efficient execution of neural network operations.
Innovation Solution
An in-memory computing (IMC) macro that can alternate between a spiking neural network (SNN) mode and an artificial neural network (ANN) mode, utilizing an input control circuit, crossbar array, and post arithmetic circuit to perform operations directly on data stored in memory, minimizing data movements and optimizing power efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data is frequently moved between processor and memory, then computation can be performed, but power efficiency deteriorates and operational performance decreases
Solution Approach 1:
The patent merges memory storage and computation functions into a single integrated structure. The crossbar array simultaneously serves as memory for storing weights and as a computation engine for performing MAC operations, eliminating the need to move data between separate processor and memory components.
Solution Approach 2:
The crossbar array is designed to perform multiple functions: storing weights in memory cells, performing parallel MAC operations across multiple columns, supporting both SNN and ANN modes, and enabling feedback operations. This multi-functionality eliminates the need for dedicated separate components for each operation type.
2Adaptability or versatility
If separate hardware is added to support multiple neural network modes, then functionality increases, but device complexity increases
Solution Approach 1:
The same crossbar array hardware structure supports both SNN and ANN modes through configurable operation modes. The input control circuit and post arithmetic circuit are configured differently depending on the desired mode, but the core computation engine remains the same, avoiding the need for separate hardware for each neural network type.
Solution Approach 2:
The system dynamically switches between SNN and ANN modes based on operational requirements. The input control circuit and post arithmetic circuit adapt their behavior according to the selected mode, allowing a single hardware structure to serve multiple purposes without requiring physical reconfiguration.
3Loss of energy
If in-memory computing is implemented, then power efficiency improves, but the ability to perform diverse operations decreases
Solution Approach 1:
The crossbar array is designed to perform both SNN operations (spike processing with thresholding) and ANN operations (standard MAC with activation functions) using the same hardware structure. The input control circuit configures the operation mode to match the desired computation type, maintaining power efficiency while providing operational diversity.
Solution Approach 2:
The system changes operational parameters (such as the nature of input signals, threshold values, and post-processing operations) to support different neural network modes. By adjusting these parameters rather than changing the hardware structure, the system maintains power efficiency while achieving operational versatility.
Data Source
AI summary
An in-memory computing (IMC) macro has a mode alternating between a first mode and a second mode, and the IMC macro includes: an input control circuit configured to be capable of generating a signal in which a predefined pattern is applied to an input signal and of transmitting a previous operation result that is fed back, and which is performed depends on which mode the operating mode is in; a crossbar array including memory cells including an additional row that processes and stores the fed-back previous operation result, and columns including an adder tree corresponding to the memory cells; and a post arithmetic circuit configured to be capable of performing a first operation corresponding to a spiking neural network (SNN) and a second operation corresponding to an artificial neural network (ANN), wherein which of the first and second operations is performed depends on which mode is in effect.