Discrete-Time Analog Filtering for Accurate RRAM MAC Operations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Resistive random-access memory (RRAM) cells face limitations in controlling written values, which affects their accuracy in analog computations, leading to errors that are challenging to manage, especially when sampling and storing analog signals.
Innovation Solution
A discrete-time analog filtering apparatus and method that employs oversampling and noise shaping to increase accuracy in RRAM cells, allowing for efficient and accurate Multiply-Accumulate (MAC) operations by displacing quantization noise from the baseband into a wider bandwidth, where it can be attenuated, thereby improving signal-to-noise ratio and maintaining accuracy throughout computations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If RRAM cells are used to store analog values for MAC operations, then computational speed and energy efficiency are improved, but manufacturing precision and control over written values deteriorate
Solution Approach 1:
The patent implements a feedback mechanism where the actual written value in the RRAM cell is read back and compared with the intended value. The difference (error) is calculated and used to adjust subsequent write operations, enabling iterative convergence to the target value. This closed-loop feedback system compensates for the poor write control characteristics of RRAM cells.
Solution Approach 2:
The patent replaces direct analog voltage writing with a digital feedback-controlled writing mechanism. Instead of relying on analog voltage precision during writing, the system uses digital error calculation and iterative adjustment, substituting the mechanical/analog writing process with a digital control system that is better suited to RRAM's characteristics.
2Device complexity
If analog values are directly stored in RRAM cells, then device complexity is reduced, but measurement precision and computational accuracy deteriorate
Solution Approach 1:
The patent segments the storage and computation process into discrete steps: storing kernel values in RRAM cells, reading them back, calculating errors against intended values, and iteratively adjusting. This segmentation allows the system to use simple RRAM storage while maintaining high precision through multiple discrete correction steps.
Solution Approach 2:
The patent performs preliminary actions by pre-calculating the expected kernel values and storing them for comparison. Before finalizing the MAC operation, the system pre-computes error terms and prepares correction values, ensuring that precision-critical calculations are performed with accurate data.
3Measurement precision
If error correction mechanisms are implemented in RRAM-based MAC operations, then computational accuracy is improved, but device complexity and operational overhead increase
Solution Approach 1:
The patent merges the error correction process with the MAC operation itself. The read-back, error calculation, and value adjustment are integrated into the same computational flow rather than being separate correction phases. This merging reduces overhead by combining multiple functions into a unified process.
Solution Approach 2:
The system implements self-service error correction where the RRAM cell's own read output is used to generate the correction signal. The cell essentially corrects itself by providing feedback about its actual state, eliminating the need for external complex correction circuits.
Data Source
AI summary
According to an example, discrete-time analog filtering may include receiving an input signal, and sampling the input signal to determine sampled input signal values related to the input signal.