GPU State Error Injection for Precise Soft Error Assessment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for assessing the resilience of graphics processing units (GPUs) to soft errors, such as ionizing radiation, lack control over which memory circuits are affected and when, providing an inaccurate assessment of resilience during application execution.
Innovation Solution
A method for precise state error injection in GPUs, allowing for the emulation of soft errors by halting execution in a specified subsystem, injecting a state error at a specified data storage circuit or RAM bit, and resuming execution, enabling transparent assessment of error consequences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If ionizing radiation is used to subject GPU to soft errors, then resilience assessment can be performed, but control over which memory circuits are affected and when is lost
Solution Approach 1:
The patent introduces an intermediary error injection mechanism that sits between the radiation source and the GPU memory circuits. This intermediary system allows controlled injection of soft errors into specific memory circuits at predetermined times, combining the realism of radiation-induced errors with the controllability of a programmable system. The intermediary translates high-level error injection commands into precise physical error injections at the circuit level.
Solution Approach 2:
The patent replaces the uncontrollable physical process of ionizing radiation with a controllable electronic error injection mechanism. Instead of relying on random radiation events, the system uses electronic circuits to deterministically inject errors into specific memory locations at specific times, substituting a mechanical/physical process with an electronic control system that offers precise programmability and repeatability.
2Measurement precision
If conventional resilience testing is performed without targeted error injection, then testing is simpler, but assessment accuracy for specific applications is insufficient
Solution Approach 1:
The patent segments the GPU memory system into individually addressable units (flip-flops, latches, RAM bits) that can be targeted for error injection. By dividing the memory system into discrete controllable elements, the patent enables precise error injection into specific segments rather than requiring system-wide radiation exposure, thereby improving measurement precision while managing complexity through modular addressing.
Solution Approach 2:
The patent implements preliminary action by pre-configuring error injection parameters (target memory circuit, error type, injection timing) before actual error injection occurs. This allows the system to prepare test scenarios in advance, ensuring that errors are injected at critical points in application execution, thereby improving assessment accuracy without requiring complex real-time control during error injection.
Data Source
AI summary
Unavoidable physical phenomena, such as an alpha particle strikes, can cause soft errors in integrated circuits. Materials that emit alpha particles are ubiquitous, and higher energy cosmic particles penetrate the atmosphere and also cause soft errors. Some soft errors have no consequence, but others can cause an integrated circuit to malfunction. In some applications (e.g. driverless cars), proper operation of integrated circuits is critical to human life and safety. To minimize or eliminate the likelihood of a soft error becoming a serious malfunction, detailed assessment of individual potential soft errors and subsequent processor behavior is necessary. Embodiments of the present disclosure facilitate emulating a plurality of different, specific soft errors. Resilience may be assessed over the plurality of soft errors and application code may be advantageously engineered to improve resilience. Normal processor execution is halted to inject a given state error through a scan chain, and execution is subsequently resumed.