Mixed-Mode Crossbar Array for In-Memory Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computing architectures, such as von Neumann architectures, face limitations in speed and energy efficiency due to the 'Von-Neumann bottleneck,' especially in data-centric applications like machine learning, where data movement dominates energy consumption and limits operating speed, and existing custom hardware like GPUs and FPGAs are constrained by Joule heating, RF crosstalk, and capacitance.
Innovation Solution
A mixed-mode crossbar array that integrates optical and electrical interactions using mixed-mode memory elements, allowing both optical and electrical access, with programmable states that can be manipulated optically and read electrically, enabling efficient matrix-vector multiplication and neural network training through optical and electrical readout signals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If data is transported between memory and processor in von Neumann architecture, then data processing can be performed, but energy consumption increases and operating speed decreases due to the Von-Neumann bottleneck
Solution Approach 1:
The patent merges memory and processing functions into a single integrated structure where memory elements directly perform computational operations. The crossbar array integrates storage and processing resources, eliminating the need for separate data transport between memory and processor, thus resolving the energy-speed tradeoff by making processing occur at the memory location.
2Productivity
If custom hardware like GPUs and FPGAs is used to accelerate MAC computations, then computing capability increases, but the hardware is constrained by Joule heating, RF crosstalk, and capacitance
Solution Approach 1:
The patent replaces electronic computation mechanisms with optical mechanisms. Light-based operations in the photonic crossbar array eliminate the electrical resistance and capacitive effects that cause Joule heating and RF crosstalk in electronic hardware, while maintaining high computing capability through optical signal processing.
3Productivity
If electronic components are used for neural network hardware accelerators, then logic gate efficiency can be improved, but data movement still requires charging and discharging of interconnects, limiting maximum operating speed
Solution Approach 1:
The patent substitutes optical signals for electrical signals in data movement and processing. Optical signals propagate through waveguides without requiring charging and discharging of capacitive interconnects, enabling faster operation speeds while maintaining efficient computation through optical logic operations in the crossbar array.
4Use of energy by moving object
If in-memory computing is implemented using non-volatile memory crossbar arrays, then power consumption decreases and parallelism increases, but device reliability and variability remain challenges
Solution Approach 1:
The patent replaces electronic memory devices with photonic memory elements that use optical properties (such as phase, amplitude, or wavelength of light) to store and process information. This substitution maintains the in-memory computing paradigm's low power consumption and high parallelism while potentially improving reliability by avoiding the variability issues inherent in electronic non-volatile memory devices.
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 reduces power consumption and latency by performing computations at the site of data storage, enhancing computational throughput and energy efficiency, and supports parallel operations, overcoming the limitations of traditional electronic components.
Implementation Method 1
The mixed-mode memory element is configured to manipulate the portion of the input optical signal according to the optical property of the persistent state to produce an output
Implementation Method 2
An electrical readout circuit is configured to electrically interact with the mixed-mode memory element to produce an electrical readout signal that varies with the electrical property of the mixed-mode memory element
Implementation Method 3
The input optical coupler is operatively coupled to the input optical waveguide and configured to couple at least a portion of the input optical signal onto an optical processing pathway
Data Source
AI summary
Embodiments of the present disclosure generally provide for a method and apparatus for performing computations using mixed-mode memory elements having contents which are accessible via both optical and electrical interactions. In particular, an array of such elements are accessed through a “crossbar” array structure which includes both an optical crossbar array structure an electrical crossbar array structure. Applications to machine learning, e.g. neural network training, are also provided for.