Electrostatic MEMS Neurons for Low-Power CTRNN Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current digital silicon technology and von Neumann architecture face challenges in processing large data sets due to high power consumption and complex thermal management, making them unsuitable for real-time implementation in applications like wearable devices, where continuous-time recurrent neural networks (CTRNNs) are computationally expensive due to the need for simultaneous solutions of highly-coupled differential equations.
Innovation Solution
A CTRNN is implemented using micro-electro-mechanical system (MEMS) devices that exploit nonlinear dynamics and bi-stability to simulate neurons, allowing for interconnected MEMS devices to perform sensing and processing tasks directly, reducing power consumption and increasing efficiency by using AC or DC voltages to achieve bi-stable behavior.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If CMOS-based logic gates are used for digital computing, then computational performance and scalability are improved, but power consumption increases and thermal management becomes complicated
Solution Approach 1:
The patent replaces electronic CMOS-based logic gates with mechanical MEMS devices that utilize physical phenomena (electrostatic forces, nonlinear dynamics, bi-stability) to perform computational functions. This substitution fundamentally changes the computing paradigm from digital electronics to mechanical systems with inherent memory capabilities, thereby reducing power consumption while maintaining computational performance.
Solution Approach 2:
The patent changes the operating parameters from digital voltage levels to mechanical displacement states (two stable positions). By utilizing the bi-stable characteristic of MEMS devices where the proof mass can reside in either of two stable positions, the system achieves memory functionality without continuous power supply, dramatically reducing power consumption compared to CMOS logic that requires constant power for state maintenance.
2Adaptability or versatility
If CTRNNs are implemented using traditional computing methods, then neural network functionality is achieved, but computational cost increases due to simultaneous solutions of highly-coupled differential equations
Solution Approach 1:
The patent implements CTRNNs using interconnected MEMS devices where each device's nonlinear dynamics and bi-stability inherently solve the differential equations without external computational assistance. The system is self-sufficient, utilizing the natural physical behavior of MEMS devices to perform the mathematical operations required for neural network functionality, thereby eliminating the need for expensive simultaneous equation solvers.
Solution Approach 2:
The patent exploits the dynamic behavior of MEMS devices, particularly their nonlinear dynamics and transition between stable states, to simulate neural network activation functions. By configuring MEMS devices to operate in their nonlinear regime, the system naturally exhibits behavior analogous to neural activation, enabling CTRNN implementation without explicit numerical solution of differential equations.
3Use of energy by moving object
If MEMS devices are used to simulate neurons, then power consumption is reduced and processing speed increases, but device complexity increases due to nonlinear dynamics requirements
Solution Approach 1:
The patent changes the operational parameters of MEMS devices from linear elastic regime to nonlinear regime by applying appropriate bias voltages and designing the proof mass-spring system to operate near instability points. This parameter change enables bi-stable behavior which provides memory functionality, accepting increased device complexity as a trade-off for dramatically reduced power consumption.
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
This approach enables faster and more power-efficient processing within MEMS devices, capable of performing complex tasks similar to traditional CTRNNs but with significantly reduced power consumption and increased speed, making them suitable for applications with limited resources.
Implementation Method 1
The MEMS device is associated with non-linear dynamics that cause the MEMS device to exhibit a bi-stable response to an input signal
Implementation Method 2
Each MEMS device in the CTRNN is configured to simulate a neuron by exploiting the characteristics of bi-stability and hysteresis inherent in certain MEMS device structures
Implementation Method 3
Each MEMS device in the CTRNN is configured to simulate a neuron by exploiting the characteristics of bi-stability and hysteresis inherent in certain MEMS device structures
Implementation Method 4
A force acting on the proof mass of the MEMS device changes a displacement of the microbeam
Data Source
AI summary
A continuous-time recurrent neural network (CTRNN) is described that exploits the nonlinear dynamics of micro-electro-mechanical system (MEMS) devices to model a neuron in accordance with a neuron rate model that is the basis for dynamic field theory. Each MEMS device in the CTRNN is configured to simulate a neuron population by exploiting the characteristics of bi-stability and hysteresis inherent in certain MEMS device structures. In an embodiment, the MEMS device is a microbeam or cantilevered microbeam device that is excited with an alternating current (AC) voltage at or near an electrical resonance frequency associated with the MEMS device. In another embodiment, the MEMS device is an arched microbeam device that is excited with a direct current voltage and exhibits snap-through behavior due to the physical design of the structure. A CTRNN can be implemented using a number of MEMS devices that are interconnected, the connections associated with varying connection coefficients.