Adaptive Dynamic Voltage Control for Near-Threshold Energy Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing Near/Sub-threshold technology is limited by low speed performance and process variation, which hinders its practical application in commercial chips due to excessive area and power overhead, and lacks optimization for specific power per performance requirements.
Innovation Solution
A dynamic and adaptive operating voltage control system that samples process corners and temperature on the fly, combined with a programmable DC2DC converter and first fail circuits, to optimize energy per operation while ensuring target performance, using sensors and a software programmable algorithm to determine the optimal voltage levels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If operating voltage is reduced to Near/Sub-threshold levels, then power consumption is dramatically reduced, but speed performance deteriorates exponentially
Solution Approach 1:
The patent implements dynamic voltage scaling that adjusts operating voltage in real-time based on actual workload and performance requirements. Instead of statically fixing voltage at Near/Sub-threshold levels, the system dynamically transitions between voltage domains (Near-threshold, Sub-threshold, and higher) to match instantaneous computational demands, thereby maintaining acceptable speed performance while minimizing power consumption during low-utilization periods
Solution Approach 2:
The system changes the operating voltage parameter adaptively based on process corner measurements and temperature conditions. By characterizing the actual silicon device parameters (process variations) and adjusting voltage accordingly, the system optimizes the trade-off between speed and power for each specific device instance, rather than designing for worst-case conditions across all devices
2Reliability
If designing for worst case process conditions, then reliability is improved, but area and power overhead increase dramatically
Solution Approach 1:
The patent applies different operating conditions to different devices based on their individual process corner characteristics. Each device is characterized during manufacturing to determine its specific process corner (fast, typical, slow), and then configured to operate at the appropriate voltage level for that corner. This allows fast devices to operate at lower voltages for power savings while slow devices operate at higher voltages to meet performance targets, eliminating the need to design all devices for worst-case conditions
Solution Approach 2:
The system performs preliminary characterization of process corners and device parameters during manufacturing or initial operation. Based on these pre-measured characteristics, the system pre-configures optimal voltage settings and performance targets for each device before actual workloads are executed. This preliminary action eliminates the need for conservative worst-case design margins, as each device is already optimized for its specific characteristics
3Use of energy by moving object
If using adaptive dynamic voltage control, then energy per operation is optimized, but device complexity and control overhead increase
Solution Approach 1:
The system implements self-service through automated feedback loops that monitor actual performance metrics (throughput, latency, power consumption) and autonomously adjust voltage settings without external intervention. The control system reads performance counters, compares against targets, and automatically scales voltage to maintain optimal energy-per-operation ratios. This self-regulating mechanism reduces the need for complex external control infrastructure while achieving sophisticated power optimization
Solution Approach 2:
The patent incorporates multiple feedback mechanisms including performance counter feedback (monitoring operations per cycle), power consumption feedback (measuring actual power draw), and temperature feedback (monitoring thermal conditions). These feedback signals are continuously fed into the voltage control logic, which adjusts operating voltage in response to measured conditions. This closed-loop feedback system enables the complex energy optimization to occur automatically based on real-time system state rather than pre-computed static configurations
Data Source
AI summary
A method for implementing a Semiconductor Integrated Circuit device using Near/Sub-threshold technology with SOFTWARE programmable Adaptive Dynamic Voltage Control (ADVC) algorithm using different sensors inside the chip in order to improve the target speed and reduce the energy per operation of the final product. This method achieves the best power per performance for a given solution operating at a required speed.