Fault-Tolerant Processing Unit with Sub-Unit Segmentation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Highly integrated hardware accelerators used in AI and neural networks are prone to hardware faults due to small structure sizes, leading to potential failures in computationally intensive tasks.

Innovation Solution

A method for fault-tolerant operation of processing units with multiple sub-processing units, where adaptation partwords are determined from input words to enable continued operation even if a fault occurs in a sub-processing unit, by processing these adaptation partwords in a second operation mode with reduced accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Use of energy by moving object

If integrated circuits are designed with the smallest possible structure sizes to reduce energy consumption and costs, then energy consumption and costs are reduced, but the circuits become more susceptible to hardware faults

Engineering Contradiction:

Improveenergy consumptionVSAvoidsusceptibility to hardware faults

Core Design Contradiction:

Use of energy by moving objectVSReliability

Solution Approach 1:

The processing unit is divided into multiple sub-processing units, each capable of independently processing partwords. This segmentation allows the system to continue operating with a subset of functional sub-units even when faults occur, thereby maintaining reliability while keeping individual structure sizes small for low energy consumption.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system performs preliminary actions by determining adaptation partwords from input words before processing, and by having spare sub-processing units ready. When faults are detected, the system can immediately switch to using adaptation input words processed by remaining functional sub-units, ensuring continuous operation without complete failure.

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If the processing unit operates in the first operation mode with full input words, then processing accuracy is maintained, but the system fails completely when a fault occurs in a sub-processing unit

Engineering Contradiction:

Improveprocessing accuracyVSAvoidsystem continuity upon fault

Core Design Contradiction:

Measurement precisionVSReliability

Solution Approach 1:

The processing unit dynamically switches between two operation modes: the first operation mode processes full input words for high accuracy when all sub-units are functional, while the second operation mode processes adaptation input words using fewer sub-units when faults occur. This dynamic adaptation maintains system continuity while adjusting accuracy based on operational conditions.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system changes operational parameters by switching between different operation modes. When faults are detected, parameters such as input word format (full vs. adaptation), processing precision, and the number of active sub-processing units are adjusted. This allows the system to continue operating with modified parameters rather than complete failure.

Inventive Principle:

Principle #35Parameter changes

3Reliability

If adaptation input words are processed in the second operation mode using fewer sub-processing units, then continued operation is enabled with reduced accuracy, but complete processing capability is lost

Engineering Contradiction:

Improvecontinued operation capabilityVSAvoidprocessing accuracy

Core Design Contradiction:

ReliabilityVSMeasurement precision

Solution Approach 1:

The system performs partial processing by using only the functional sub-processing units available after faults occur, rather than requiring all units to operate. Adaptation input words are designed to be processable by a subset of sub-units, enabling partial but sufficient operation to prevent complete system failure, even though full processing accuracy cannot be maintained.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS20250077363A1Method for fault-tolerant operation of a processing unit and of a processing arrangement, circuit arrangement, and computing unit

Publication Date: 2025.03.06 ROBERT BOSCH GMBH

AI summary

A method for fault-tolerant operation of a processing unit having a plurality of sub-processing units. Each sub-processing unit is configured to process one or more partwords and to determine a result partword therefrom. The processing unit can be configured in a plurality of operation modes and is configured to process partwords of input words, which are arranged in sections of the input words, in the sub-processing units corresponding to the sections, and to determine an output word from the result partwords determined thereby depending on the operation mode. The method includes: determining adaptation partwords from input words to be processed that are suitable for processing in at least one fault-free sub-processing unit, and forming adaptation input words, in which the adaptation partwords are arranged in sections corresponding to the fault-free sub-processing unit; and processing the adaptation input words in a second operation mode to determine an adaptation output word.