Neuromorphic Unit With Orthogonal Spikes for Parallel Workloads
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neuromorphic chips face limitations in managing multiple independent workloads simultaneously, leading to inefficient processing and reduced data-carrying capacity due to the need to handle each spike separately, which is not addressed by existing approaches that embed workload IDs within spike payloads.
Innovation Solution
Implementing modulated spikes, specifically orthogonally modulated spikes, in neuromorphic units to enable parallel DNN workloads by using Code Division Multiple Access (CDMA) techniques and orthogonal coding to differentiate between tasks, allowing multiple spike streams to coexist without interference.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If workload IDs are embedded within spike payloads, then workload identification is achieved, but data-carrying capacity is reduced and processing efficiency deteriorates
Solution Approach 1:
The patent segments the spike signal into distinct temporal patterns rather than embedding workload IDs within the payload. Each workload is assigned a unique spike pattern (e.g., different frequencies or rhythms), allowing workload identification through temporal segmentation of the spike stream, thereby preserving full payload capacity for data transmission.
Solution Approach 2:
The patent introduces temporal modulation as an intermediary mechanism between the workload identifier and the spike payload. Instead of directly embedding IDs in the payload, the system uses temporal patterns as a mediator that carries workload information separately from the data, allowing both efficient data transmission and reliable workload identification.
2Reliability
If multiple spike streams are handled separately, then workload isolation is maintained, but processing complexity increases and hardware efficiency decreases
Solution Approach 1:
The patent merges multiple spike streams into a single unified stream by multiplexing them using orthogonal temporal patterns. Different workloads are represented by distinct temporal codes (such as different frequencies or phase patterns) that can be simultaneously encoded in one spike stream, allowing the hardware to process all workloads through a single unified processing path without increasing complexity.
Solution Approach 2:
The patent creates a universal spike stream that can carry information from multiple workloads simultaneously through temporal coding. A single spike stream serves multiple functions by encoding different workload identifiers through temporal patterns, eliminating the need for separate processing paths for each workload and thereby reducing hardware complexity.
3Device complexity
If sequential processing is used, then workload management is simplified, but processing time increases and productivity decreases
Solution Approach 1:
The patent uses periodic temporal patterns (such as different frequencies or rhythmic patterns) to encode workload identifiers in spike streams. These periodic patterns allow multiple workloads to be processed in parallel through time-division multiplexing, where each workload is assigned a distinct temporal period or frequency, enabling simultaneous processing without increasing management complexity.
Solution Approach 2:
The patent assigns temporal codes and patterns to workloads in advance before processing begins. By pre-configuring the temporal encoding scheme for different workloads, the system enables parallel processing without requiring complex runtime scheduling decisions, thereby maintaining simple workload management while significantly improving throughput.
Data Source
AI summary
An apparatus may have a neuromorphic architecture and facilitate parallel neural network workloads. The apparatus may include neurons and connections between the neurons. The neurons may perform computations in the neural network and produce payloads and may transmit payloads to each other in the form of spikes. The spike messages may be orthogonally modulated spikes that facilitate executions of the neural network to process multiple inputs to the neural network through parallel workloads. A spike message may indicate a payload and identify the workload through which the payload is produced. A neuron may combine multiple spikes into a combined spike. The combined spike may represent all the payloads and workloads in the spike messages. The neuron may send the combined spike to another neuron, which may decode the combined spike to obtain a payload for performing a neural network operation for a workload that it is assigned to.