Memristive Multiplication Device for Complex MAC Operations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing memristive arrays face inefficiencies in performing complex multiply-accumulate operations due to the need for duplicate memristive elements to represent complex weight components, leading to increased latency and impracticality in signal processing and neural network applications.
Innovation Solution
A memristive multiplication device using two memristive cells, each representing a component of a complex weight value, allows for simultaneous execution of sub-computations by employing real and imaginary input multipliers to perform multiplications in parallel, eliminating the need for duplicate elements and reducing latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If duplicate memristive elements are used to represent complex weight components, then the complex MAC operation can be performed, but the device complexity and latency increase
Solution Approach 1:
The complex weight is segmented into real and imaginary components, with each component stored in separate memristive elements. This segmentation allows the complex multiplication to be decomposed into simpler real-valued operations that can be executed in parallel, reducing the overall computational complexity and latency while maintaining the capability to perform complex MAC operations.
Solution Approach 2:
The patent transitions from representing complex weights using duplicate memristive elements in the same dimension to using separate real and imaginary components that occupy different dimensional spaces in the computational architecture. This dimensional separation enables parallel processing of real and imaginary parts, reducing device complexity while preserving complex arithmetic capability.
2Adaptability or versatility
If duplicate memristive elements are used to represent complex weight components, then the complex MAC operation can be performed, but the processing speed decreases
Solution Approach 1:
By segmenting the complex weight into real and imaginary components stored in separate memristive elements, the computation is divided into parallel sub-operations. This segmentation enables simultaneous processing of real and imaginary parts, significantly improving processing speed compared to sequential operations required by duplicate element approaches.
Solution Approach 2:
The patent maintains continuous useful action by enabling parallel execution of real and imaginary component multiplications. Instead of alternating between duplicate elements, the system performs both real and imaginary computations simultaneously, eliminating idle time and maintaining continuous productive operation throughout the MAC computation.
3Adaptability or versatility
If duplicate memristive elements are used to represent complex weight components, then the complex MAC operation can be performed, but the latency increases
Solution Approach 1:
Segmenting the complex weight representation into separate real and imaginary components allows the multiplication operation to be divided into parallel sub-operations. This segmentation eliminates the sequential dependencies inherent in duplicate element approaches, reducing the critical path and overall latency of the complex MAC operation.
Solution Approach 2:
The patent performs preliminary action by pre-separating and storing real and imaginary weight components in dedicated memristive elements before the MAC operation begins. This preliminary organization of data enables immediate parallel computation when the operation is triggered, eliminating the need for runtime element switching or duplication that causes latency in conventional approaches.
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 enhances the efficiency of complex MAC operations by enabling fast, real-time processing of complex-valued signals and weights, improving the performance of filter and neural network operations without the need for redundant memristive elements.
Implementation Method 1
Each memristive cell includes a memristive element to store a component of a complex weight value
Data Source
AI summary
In one example in accordance with the present disclosure a device is described. The device includes at least two memristive cells. Each memristive cell includes a memristive element to store one component of a complex weight value. The device also includes a real input multiplier coupled to the memristive element to multiply an output signal of the memristive element with a real component of an input signal. An imaginary input multiplier of the device is coupled to the memristive element to multiply the output signal of the memristive element with an imaginary component of the input signal.