Hybrid Optical Digital Computing Workload Scheduling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data-center chips face thermal limitations due to high power consumption, especially with increasing transistor density, leading to inefficiencies in silicon space utilization and the 'dark silicon' challenge, where not all transistors can be powered simultaneously.
Innovation Solution
A hybrid computing system that combines digital and optical computing units, using silicon photonics to determine the most efficient domain for workload execution based on performance measures, and optimizing power consumption by reducing overhead in digital-to-analog and analog-to-digital converter efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If transistor density is increased to improve computational power, then processing capability is improved, but power consumption increases leading to thermal limits
Solution Approach 1:
The system segments computational tasks into two distinct domains: digital computing for control logic and data processing, and optical computing for matrix multiplications and data-intensive workloads. This segmentation allows each domain to handle tasks most suitable to its characteristics, with optical computing handling power-intensive operations while digital computing manages overhead, thereby improving overall power efficiency while maintaining computational power
Solution Approach 2:
The patent replaces traditional electronic computing mechanisms with optical computing mechanisms for specific computational tasks. By using photonic-based computing arrays instead of transistor-based electronic circuits for matrix multiplications and data-intensive workloads, the system achieves higher computational power with significantly reduced power consumption and heat generation, directly addressing the thermal limits problem
2Productivity
If more chip cards are placed in a rack unit to increase system capacity, then system throughput is improved, but thermal density increases causing cooling challenges
Solution Approach 1:
By substituting electronic computing with optical computing for data-intensive workloads, the patent dramatically reduces the thermal output per chip. Optical computing arrays consume significantly less power and generate far less heat compared to traditional high-density transistor-based chips, enabling higher system throughput through increased chip density without proportionally increasing thermal density and cooling challenges
3Use of energy by moving object
If optical computing is introduced to reduce power consumption, then power efficiency is improved, but system complexity increases due to domain conversion requirements
Solution Approach 1:
The patent merges digital and optical computing domains into a unified hybrid system where both domains work together through standardized interfaces. The digital computing unit handles control logic, data preparation, and result processing, while the optical computing unit handles matrix multiplications and data-intensive operations. This merging allows the system to leverage the power efficiency of optical computing while maintaining the flexibility and ease of control of digital computing, offsetting the added complexity through integrated architecture design
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 power efficiency by selectively scheduling tasks between digital and optical domains, allowing for scalable performance without significant power increase, thereby maximizing silicon space utilization and reducing thermal constraints.
Implementation Method 1
optical computing units are proposed that apply silicon photonics to process data-intensive workloads
Data Source
AI summary
Methods and systems are disclosed for reducing power consumption by a system including a digital unit and an optical unit. Techniques disclosed comprise generating a workload signature of an incoming workload to be executed by the system. Based on the generated workload signature, techniques disclosed comprise matching the incoming workload with a profile of stored workload profiles. The workload profiles are generated by a trace capture unit. Based on the associated profile, a task submission transaction is sent to the optical unit of the system, representative of a request to execute the incoming workload by the optical unit.