Systolic Array Operator Pre-Operation for IR Drop Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Systolic arrays face reliability issues due to rapid increases in power consumption and voltage instability (IR drop) when many arithmetic operators start operating simultaneously, leading to errors and delayed execution timing, which compromises processing performance.
Innovation Solution
Implementing idling control for arithmetic operators by pre-operating them with power before actual use, limiting the number of operators starting operations to avoid IR drop and ensure stable voltage, thereby maintaining performance and reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If many arithmetic operators start operating simultaneously to increase processing speed, then productivity is improved, but voltage stability deteriorates due to IR drop
Solution Approach 1:
The control unit pre-activates arithmetic operators before they are actually needed for computation. By calculating which operators will be needed in future cycles and activating them in advance, the system prepares the operators during idle periods rather than activating them all at once when computation begins, thus avoiding sudden power surges and voltage drops
Solution Approach 2:
The system activates arithmetic operators in a periodic, staged manner rather than all at once. The control unit divides operator activation into multiple phases, activating a subset of operators in each cycle based on predicted computation needs, which distributes the power consumption over time and prevents simultaneous activation of all operators
2Productivity
If the number of arithmetic operators is increased to handle more arithmetic operations, then productivity is improved, but power consumption increases causing heat and voltage instability
Solution Approach 1:
The control unit predicts which arithmetic operators will be needed based on the computation task and activates them in advance during idle periods. This allows the system to have a large number of operators available when needed while only actually powering them up when required, rather than keeping all operators continuously powered
Solution Approach 2:
The system dynamically adjusts the activation state of arithmetic operators based on real-time computation needs. The control unit continuously monitors which operators are currently being used and which will be needed next, activating or deactivating operators accordingly to match the actual computational workload rather than maintaining a static configuration
3Reliability
If arithmetic operators are activated in advance to prevent voltage drops, then reliability is improved, but loss of time occurs during pre-operation
Solution Approach 1:
The control unit performs preliminary activation of arithmetic operators during idle periods before the actual computation begins. By analyzing the computation task in advance, the system determines which operators will be needed and activates them during non-computation time, so no additional time is lost during the actual computation phase
Solution Approach 2:
The control unit uses feedback from the computation task analysis to optimize operator activation timing. By monitoring which operators are currently active and predicting future needs, the system adjusts activation schedules to minimize idle pre-operation time while ensuring operators are ready when needed, balancing preparation time against computation efficiency
Data Source
AI summary
An arithmetic device includes: a memory configured to receive and store data for a matrix arithmetic operation including individual arithmetic operations; a systolic array in which arithmetic operators are arranged in a matrix, and configured to execute the matrix arithmetic operation by causing the arithmetic operators to perform different individual arithmetic operations; and a processor configured to, upon receiving a processing request for the matrix arithmetic operation, calculate an excess arithmetic operator number which is a number of arithmetic operators that exceed an upper limit number to be caused to newly start operation among increments of the arithmetic operators to be caused to perform the individual arithmetic operations, pre-operate the arithmetic operators based on the excess arithmetic operator number based on start of data reception of the memory, and cause the systolic array to execute the matrix arithmetic operation based on the data when the memory completes the data reception.