Neural Network Processor Window Expander Circuit
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neural network architectures struggle to efficiently compute both Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) simultaneously, often specializing in one type and compromising on performance for the other, while also facing challenges with real-time latency requirements.
Innovation Solution
A neural network processor is designed with a window expander circuit that expands input data to increase utilization of multiplier circuits, leveraging parallelism between output activations in CNNs and using a CNN-aware instruction set architecture to perform efficient matrix-matrix multiplication, allowing for high throughput at low latency for both CNNs and RNNs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If neural network processor specializes in one type of network (CNN or RNN), then performance for that type is improved, but performance for the other type deteriorates
Solution Approach 1:
The patent implements a universal neural network processor that can efficiently handle both CNN and RNN operations through a unified architecture. The window expander circuit and multiplier circuits are designed to support multiple network types, allowing the same hardware to perform both convolutional and recurrent operations without specialization, thus achieving high productivity for both network types simultaneously
2Productivity
If input data is expanded to increase multiplier circuit utilization, then processing throughput is improved, but computational complexity increases
Solution Approach 1:
The window expander circuit divides the input data into multiple segments and processes them through parallel multiplier circuits. By segmenting the data processing task and distributing it across multiple multiplier circuits, the system achieves high throughput without requiring a single overly complex circuit, thus balancing productivity gains with manageable device complexity
3Loss of time
If real-time processing is implemented to meet latency requirements, then response time is improved, but processing accuracy may deteriorate
Solution Approach 1:
The window expander circuit performs preliminary data expansion and preparation before the main computation occurs. By pre-processing the input data in the window expander, the system reduces the computational burden during the critical real-time processing phase, enabling both low latency and high accuracy to be achieved simultaneously through staged processing
Data Source
AI summary
Neural network processors including a window expander circuit and related methods are provided. The window expander circuit may include a first logic circuit configured to store a set of data elements, corresponding to at least a subset of the input data, into a Q number of logical memories, where each of a P number of data elements of the set of the data elements is stored in each of the Q number of logical memories. The window expander circuit may further include a second logic circuit configured to receive the first set of data elements and additional data elements corresponding to the at least the subset of the input data from the Q number of logical memories and expand the at least the subset of the input data until the at least the subset of the input data is expanded based on a predetermined factor.