ML Accelerator LBIST Scheduling for Non-Interruptive Fault Testing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning networks (MLNs) implemented in edge devices face resource limitations, and performing built-in self-tests (BIST) at run-time can disrupt operational continuity, especially in safety-critical applications.
Innovation Solution
A machine learning accelerator (MLA) with logic built-in self-test (LBIST) circuitry performs tests during idle times of statically scheduled instructions, ensuring operational continuity by using LBIST circuitry embedded in compute elements that are self-sufficient and controlled by an LBIST manager.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If built-in self-tests are performed at run-time, then reliability is improved, but operational continuity is disrupted
Solution Approach 1:
The patent segments the compute elements into those with and without LBIST circuitry, allowing selective testing. Only specific compute elements undergo testing during idle periods while others continue computation, dividing the system into test and operational segments to maintain continuity.
Solution Approach 2:
The patent performs LBIST tests during idle periods between computation blocks, preparing the tested compute elements for reliable operation before the next computation phase. This preliminary testing ensures reliability is established before the main operation begins.
2Device complexity
If compute elements are simplified to conserve resources, then device complexity is reduced, but reliability deteriorates
Solution Approach 1:
The patent extracts the LBIST circuitry from the main compute element structure, placing it as a separate, optional component. This allows compute elements to be simplified for resource-constrained edge devices while still providing fault detection capability through the extracted test circuitry when needed.
Solution Approach 2:
The LBIST circuitry performs self-testing of the compute element without requiring external test equipment or interrupting normal operation. The circuitry serves itself by automatically detecting faults within its own structure during idle periods, enabling simplified compute elements to maintain reliability.
3Reliability
If LBIST tests are performed during computation, then reliability is improved, but productivity decreases
Solution Approach 1:
The patent implements periodic LBIST testing during idle periods between computation blocks rather than continuously during computation. This periodic action ensures reliability is maintained at appropriate intervals while allowing maximum productivity during active computation phases.
Solution Approach 2:
The patent maintains continuous useful action by performing computation during active phases and testing during idle phases, ensuring the system is always in a useful state (either computing or preparing for computation). This eliminates downtime while maintaining reliability through periodic testing.
Data Source
AI summary
Run-time logic built-in self-test (LBIST) may be performed, while ensuring operational continuity. The compute elements in a machine learning accelerator contain LBIST circuitry that performs logic testing of the functional circuitry in the compute element. The LBIST circuitry may be self-sufficient, meaning that it contains the data and instructions needed to run and evaluate these tests. An LBIST manager enables the logic testing during idle time of the functional circuitry between blocks of statically scheduled instructions. As a result, the LBIST circuitry can perform the logic tests without disrupting the computation of the machine learning network.