Multi-ASIC Adaptive Tuning for Per-Chip Voltage and Frequency Control

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Multi-ASIC systems face challenges in achieving optimal power and performance due to inherent variances in silicon manufacturing, leading to difficulties in achieving consistent power and performance across multiple application-specific integrated circuits (ASICs), which complicates frequency and voltage tuning.

Innovation Solution

An adaptive power and performance (PnP) tuning solution is implemented, using dynamic voltage and frequency scaling, smart power supply units, programmable fan speed control, and ASIC pass rate-based adaptive tuning to optimize frequency and voltage settings for individual ASICs, allowing for independent power and voltage management.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Use of energy by moving object

If traditional fixed frequency and voltage tuning is used in multi-ASIC systems, then device complexity is reduced, but power efficiency and performance optimization deteriorate due to inherent silicon manufacturing variances

Engineering Contradiction:
Improvepower efficiencyVSAvoidtuning system complexity
Core Design Contradiction:
Use of energy by moving objectVSDevice complexity

Solution Approach 1:

The patent divides the multi-ASIC system into individually tunable units, where each ASIC can have its own frequency and voltage settings adjusted independently based on its specific performance characteristics. This segmentation allows each ASIC to be optimized separately rather than forcing a uniform configuration across all devices.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements dynamic frequency and voltage tuning capabilities that allow the system to adaptively adjust operating parameters in real-time or near-real-time. This dynamic approach enables the system to respond to manufacturing variances and optimize power efficiency without requiring complete system redesign.

Inventive Principle:
Principle #15Dynamics

2Use of energy by moving object

If adaptive frequency and voltage tuning is implemented for each ASIC, then power efficiency improves by 5-10%, but device complexity and manufacturing cost increase

Engineering Contradiction:
Improvepower efficiencyVSAvoidmanufacturing complexity
Core Design Contradiction:
Use of energy by moving objectVSEase of manufacture

Solution Approach 1:

The patent implements self-service mechanisms where each ASIC includes built-in sensors and control logic that automatically monitor and adjust its own operating parameters. This self-service capability reduces the need for external tuning equipment and simplifies the manufacturing process while maintaining optimization benefits.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent incorporates feedback loops that continuously monitor ASIC performance metrics such as power consumption, temperature, and computational throughput. This feedback information is used to automatically adjust frequency and voltage settings, creating a closed-loop system that optimizes power efficiency without requiring manual intervention or complex manufacturing processes.

Inventive Principle:
Principle #23Feedback

3Productivity

If uniform frequency and voltage settings are applied to all ASICs, then device complexity is minimized, but performance consistency deteriorates due to silicon manufacturing variances

Engineering Contradiction:
Improveperformance consistencyVSAvoidtuning mechanism complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent applies local quality by allowing each ASIC to have customized frequency and voltage settings tailored to its specific performance characteristics. This localized tuning approach ensures that each device operates at its optimal point rather than being constrained by uniform system-wide settings, thereby improving overall performance consistency.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent utilizes parameter changes by systematically varying frequency and voltage parameters across different ASIC units based on their measured performance characteristics. This parameter optimization process enables the system to achieve consistent productivity levels across multiple devices despite inherent manufacturing variations.

Inventive Principle:
Principle #35Parameter changes

4Use of energy by moving object

If individual ASIC tuning is implemented, then power efficiency improves, but system complexity and difficulty of operation increase

Engineering Contradiction:
Improvepower efficiencyVSAvoidsystem operation simplicity
Core Design Contradiction:
Use of energy by moving objectVSEase of operation

Solution Approach 1:

Each ASIC is equipped with self-service capabilities including built-in sensors, control logic, and automatic adjustment mechanisms that enable the device to tune its own operating parameters without external intervention. This self-service approach maintains power efficiency benefits while significantly reducing operational complexity.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent implements feedback mechanisms that automatically monitor and adjust ASIC operating parameters based on real-time performance data. This automated feedback control eliminates the need for manual tuning operations while maintaining optimal power efficiency, thereby improving ease of operation.

Inventive Principle:
Principle #23Feedback

Data Source

PatentUS20240288850A1Adaptive tuning for multi-ASIC systems
Publication Date: 2024.08.29 INTEL CORP
  • US20240288850A1 patent drawing
  • US20240288850A1 patent drawing
  • US20240288850A1 patent drawing

AI summary

Various embodiments are directed to frequency and voltage tuning for systems with multiple application-specific integrated circuits (ASICs) and disclosed herein may be applied to multi-AIC systems in a variety of applications, such as high-performance computing, artificial intelligence, graphics applications, and cryptocurrency or blockchain mining functions.