Dual-portion synapse architecture for neural network power optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Artificial neural networks face high power consumption during both learning and inference processes, with existing methods not effectively optimizing power usage between these modes, leading to inefficiencies in processing systems.
Innovation Solution
The system employs a dual-portion synapse architecture, where a high-precision, low-precision synapse division allows for minimal power consumption during inference by enabling only necessary high-precision components during specific events, and using clock- or power-gated memory to manage weight storage, thereby reducing overall power usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If full-precision synapses are enabled during both learning and inference, then processing accuracy is maintained, but power consumption increases
Solution Approach 1:
The system dynamically switches between high-precision and low-precision synapse modes based on operational requirements. During learning operations, high-precision synapses are enabled for accurate weight updates, while during inference operations, the system can switch to low-precision synapses to reduce power consumption, thus making the precision level adaptive rather than static
Solution Approach 2:
The synapse structure is segmented into multiple precision levels (e.g., high-precision and low-precision portions). This segmentation allows the system to selectively enable only the necessary precision level for each operational mode, avoiding the need to maintain full precision continuously and thereby reducing overall power consumption while maintaining accuracy when needed
2Measurement precision
If high-precision synapses are always enabled, then learning accuracy is improved, but power consumption during inference increases
Solution Approach 1:
The precision level of synapses is made dynamic and operation-dependent. The system enables high-precision synapses during learning phases when accuracy is critical, and switches to low-precision synapses during inference phases where power efficiency is prioritized, thus optimizing the trade-off between accuracy and power consumption based on real-time operational needs
3Measurement precision
If full precision is maintained during learning, then training accuracy is improved, but overall energy expenditure increases
Solution Approach 1:
The synapse precision is segmented into multiple levels, allowing the system to use high-precision synapses only during learning operations where training accuracy is critical, and switch to low-precision synapses during inference operations. This segmentation enables the system to minimize energy expenditure by avoiding continuous use of high-precision components while maintaining training accuracy when needed
Data Source
AI summary
An information processing system, which includes a control system and an artificial neural network, is disclosed. The artificial neural network includes a group of neurons and a group of synapses, which includes a first portion and a second portion. The control system selects one of a group of operating modes. The group of neurons processes information. The group of synapses provide connectivity to each of the group of neurons. During a first operating mode of the group of operating modes, the first portion of the group of synapses is enabled and the second portion of the group of synapses is enabled. During a second operating mode of the group of operating modes, the first portion of the group of synapses is enabled and the second portion of the group of synapses is disabled.