STDP Learning Circuit Using Spike Time Difference Weight Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face difficulties in directly implementing spike-timing dependent plasticity (STDP) learning in semiconductor circuits, with synapse weights primarily being learned and operated via software.
Innovation Solution
A spike neural network circuit comprising a synapse array, neuron array, timer, input and output register arrays, time difference calculator, and weight update circuit, which autonomously performs STDP learning by calculating time differences between input and output spike signals to update synapse weights.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If STDP learning is implemented via software, then synapse weights can be learned and operated, but direct implementation in semiconductor circuits is difficult
Solution Approach 1:
The STDP learning circuit is segmented into distinct functional modules: timer unit for time value generation, register units for storing input and output spike times, time difference calculation unit for computing temporal differences, and weight update unit for adjusting synapse weights. This modular segmentation enables direct semiconductor implementation by breaking down the complex STDP learning process into manageable, implementable circuit components.
Solution Approach 2:
The patent introduces intermediate components such as the timer unit that generates time values and the register units that store spike occurrence times. These intermediaries facilitate the calculation of time differences between input and output spikes, enabling the weight update process without requiring complex direct circuit implementations. The time difference calculation unit acts as a mediator that translates temporal relationships into weight adjustment signals.
2Extent of automation
If synapse weights are updated based on time difference between spikes, then autonomous STDP learning is achieved, but additional circuit components are required
Solution Approach 1:
The circuit achieves autonomous STDP learning by incorporating self-service mechanisms where the timer unit automatically generates time values, the register units automatically store spike times when spikes occur, the calculation unit automatically computes time differences, and the weight update unit automatically adjusts synapse weights based on these differences. This automation eliminates the need for external software control, enabling the circuit to perform STDP learning independently.
Solution Approach 2:
The patent designs universal circuit components that perform multiple functions. For example, the timer unit not only generates time values but also triggers the storage operations in register units. The weight update unit receives inputs from multiple sources and performs both calculation and update operations. This multi-functionality reduces the overall number of dedicated components needed while achieving autonomous STDP learning.
3Measurement precision
If time difference calculation is performed for each spike pair, then accurate weight updates are achieved, but computational overhead increases
Solution Approach 1:
The patent implements preliminary action by having the timer unit continuously generate and store time values before spike events occur. When an input spike or output spike is detected, the corresponding time value is already available in the register units, eliminating the need for real-time time measurement during the critical weight update phase. This pre-computation approach maintains accuracy while improving computational efficiency.
Solution Approach 2:
The patent uses register units to create copies of time values at specific moments (when input or output spikes occur). Instead of continuously measuring and comparing times, the system captures snapshots of time values in the registers and performs calculations on these copied values. This copying mechanism enables accurate time difference calculation without requiring continuous computational overhead.
Data Source
AI summary
A spiking neural network (SNN) circuit performs Spike-Timing Dependent Plasticity (STDP) learning. A synapse array, with a plurality of synapse circuits, accumulates charge on corresponding column lines based on weight values in response to input spikes. A neuron array generates output spikes when accumulated charge exceeds a threshold potential. The circuit includes a timer and registers to record the occurrence times of both input and output spikes. A time difference calculator determines the temporal difference between these spikes. A weight update circuit then modifies the weight values of the synapse circuits based on this calculated time difference, enabling the SNN to update weights and learn efficiently.