Cross-Point Array Matched Filtering for Analog AI Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The expansion of cross-point arrays for in-memory computation in artificial neural networks is limited by noise issues, particularly in larger arrays where the scaling of current is constrained by noise levels, leading to signal degradation.
Innovation Solution
Incorporating matched filters into cross-point arrays to reduce noise by identifying and filtering out noise components based on a known input waveform, using either digital or analog bandpass filters tailored to the expected output frequencies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If the cross-point array size is increased to provide greater memory availability for AI applications, then the computational capability and memory capacity are improved, but noise issues worsen and signal degradation increases
Solution Approach 1:
Analog integrators are introduced as intermediary components that accumulate current signals from multiple RPU devices before they are read out. This integration process averages out random noise components while preserving the deterministic compute signal, effectively mediating between the noisy array and the readout circuitry. The integrator acts as a buffer that separates the computation domain from the measurement domain, allowing larger arrays to operate reliably.
Solution Approach 2:
The patent replaces digital sampling and processing with analog integration. Instead of converting signals to digital form early in the pipeline, the system uses continuous analog integration to accumulate and average signals, leveraging the physical properties of capacitors and resistors to perform noise reduction inherently in the analog domain before any digital conversion occurs.
2Power
If the current scaling is increased to support larger arrays, then the signal strength is improved, but noise levels increase proportionally and limit the maximum array size
Solution Approach 1:
The system employs periodic switching of wordlines and bitlines in a time-multiplexed fashion, where each RPU device is activated for a specific time window. During its active window, the device contributes to the integrated signal while remaining isolated from other devices. This periodic activation pattern allows the system to scale to larger arrays by distributing the total current load across time, preventing any single device from generating excessive noise while maintaining overall signal strength through accumulation.
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 use of matched filters enhances signal-to-noise ratio, allowing for accurate analog AI computations even in noisy environments, enabling larger and more reliable cross-point arrays.
Implementation Method 1
An input-signal matched filter is coupled to each of the columns to reduce noise in the current in accordance with the finite duration input voltage
Implementation Method 2
A plurality of input-signal matched filters each coupled to a column of the columns and including a custom bandpass filter are configured to find an input voltage signal having a sinusoidal form of finite duration in a noisy output current signal to reduce noise
Implementation Method 3
the RPU devices receive a finite duration input voltage on the rows and output a current on the columns
Data Source
AI summary
A cross-point array includes an array of Resistive Processing Unit (RPU) devices having rows and columns interconnected at cross-points, wherein the RPU devices receive a finite duration input voltage on the rows and output a current on the columns. An input-signal matched filter is coupled to each of the columns to reduce noise in the current in accordance with the finite duration input voltage.