SQUID Artificial Synapse Circuit for GHz Weight Pulses
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial neural networks face challenges with slow operation speeds, high power consumption, and large occupied space, particularly in semiconductor and superconductor-based synapses, which hinder their performance in applications requiring high-speed processing.
Innovation Solution
An artificial synapse circuit utilizing a SQUID ring and trigger circuit with Josephson junctions, combined with a converter circuit, enables high-speed, low-power, and compact neuromorphic processing by calculating and converting weights into quantized pulses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If semiconductor transistor or memristor based synapses are used, then the circuit can be manufactured with existing technology, but the operating speed is very slow (in the order of 10 Hz-1 kHz)
Solution Approach 1:
The patent changes the fundamental operating parameters by transitioning from semiconductor transistors to Josephson junctions, enabling operation at GHz frequencies instead of Hz-kHz ranges. This parameter change in the underlying physics enables the speed improvement while maintaining manufacturability through established superconducting fabrication techniques.
Solution Approach 2:
The patent replaces the semiconductor-based electronic system with a superconducting quantum interference device (SQUID) based system. By substituting the mechanical/electronic transistor structure with Josephson junctions operating on quantum tunneling principles, the system achieves dramatically higher operating speeds while remaining compatible with digital circuit interfaces.
2Speed
If superconductor-based artificial nerve cells are used, then high operating speeds can be achieved, but the circuits are relatively complex structures having large areas
Solution Approach 1:
The patent segments the complex superconductor-based synapse into modular SQUID ring units with trigger circuits. Each synapse is divided into discrete, repeatable components that can be systematically arranged, reducing overall circuit complexity and occupied area while maintaining GHz operating speeds.
Solution Approach 2:
The patent employs planar SQUID ring structures that utilize two-dimensional layout optimization. By arranging Josephson junctions and inductors in integrated planar configurations rather than three-dimensional stacked approaches, the design achieves compact footprints suitable for large-scale integration.
3Measurement precision
If more Josephson junctions are used in the synapse circuit, then higher precision weight calculation can be achieved, but the power consumption and occupied space increase
Solution Approach 1:
The patent implements partial weight calculation by utilizing only the necessary number of SQUID rings required to achieve the desired precision level. Rather than over-provisioning with excessive Josephson junctions, the design calculates and implements only the minimal number needed, optimizing the balance between precision and power consumption.
Solution Approach 2:
The patent employs periodic pulse triggering mechanisms where the SQUID rings are activated in sequential periods rather than continuously. This periodic action allows precise weight accumulation through controlled fluxon generation while minimizing power consumption by keeping the system in a low-power state between activation periods.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The proposed synapse circuit operates at GHz firing rates, consumes low power, and occupies minimal space, facilitating the development of high-performance neuromorphic processors suitable for autonomous vehicles and other AI applications.
Implementation Method 1
a SQUID ring (Superconducting Quantum Interference Device Loop) including a first inductor, a second inductor, a first junction and a second junction; a trigger circuit comprising an input, a third inductor paired with the said first inductor, a fourth inductor paired with the said second inductor
Implementation Method 2
a third inductor paired with the said first inductor, a fourth inductor paired with the said second inductor, a first resistor connected in series with the third inductor and the fourth inductor
Data Source
AI summary
Disclosed are two embodiments of an artificial synapse circuit that enable calculation of the weights in artificial neural networks. The first embodiment includes: a plurality of weighting circuits in which a plurality of single blocks are formed by connecting together in series, each having a SQUID ring and a trigger circuit, and a converter circuit adapted to convert the total weights calculated by the weighting circuits into quantized pulses. The second embodiment additionally includes: a positive side and a negative side; a positive lower inductor connected to the output of the weighting circuits on the positive side; a negative lower inductor connected to the output of the weighting circuits on the negative side; a positive upper inductor paired with the positive lower inductor; and a negative upper inductor paired with the negative lower inductor.