Dual SoC Architecture for ASIL-Certified Autonomous Driving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous driving systems face challenges in achieving automotive safety integrity levels (ASIL) certification due to non-deterministic inference models, making it difficult to ensure operational safety, especially in riskier conditions such as inclement weather or complex road scenarios.
Innovation Solution
A dual System on Chip (SoC) arrangement with isolated power sources and eFuses for electrical protection, along with direct memory access for health monitoring and seamless transitions between primary and backup roles, facilitates deterministic operations and ASIL grading of autonomous driving systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If non-deterministic inference models are used for autonomous driving, then the system can handle complex road scenarios and inclement weather conditions, but it cannot achieve automotive safety integrity levels (ASIL) certification
Solution Approach 1:
The system is divided into two separate SoCs: a primary SoC that performs non-deterministic machine learning inference for handling complex scenarios, and a backup SoC that performs deterministic safety monitoring for ASIL certification. This segmentation allows each component to specialize in its respective function without compromise.
Solution Approach 2:
A shared memory interface acts as an intermediary between the primary and backup SoCs, enabling deterministic communication and data exchange while maintaining the independence of each processor's operation mode.
2Reliability
If a dual SoC arrangement with health monitoring is implemented, then functional safety and ASIL grading are achieved, but system complexity increases
Solution Approach 1:
The backup SoC performs multiple functions simultaneously: it executes deterministic safety-critical algorithms, monitors the health status of the primary SoC, and manages failover operations. This consolidation of functions into a single monitoring unit reduces overall system complexity compared to having separate components for each function.
Solution Approach 2:
The shared memory serves multiple purposes: it stores sensor data for processing, holds intermediate computation results, and provides a communication interface for health monitoring between the primary and backup SoCs. This multi-functionality reduces the need for dedicated components.
3Reliability
If deterministic operations are implemented for ASIL grading, then safety integrity is improved, but the ability to handle non-deterministic machine learning inferences is reduced
Solution Approach 1:
The computational workload is segmented into two categories: non-deterministic machine learning inference executed by the primary SoC for handling complex scenarios, and deterministic safety-critical operations executed by the backup SoC for ensuring ASIL compliance. Each processor type is optimized for its specific operational requirements.
Solution Approach 2:
The backup SoC maintains a copy of the critical safety algorithms in a deterministic form, allowing it to independently verify and monitor the outputs of the primary SoC's machine learning models without requiring the primary system to be deterministic.
Data Source
AI summary
A system on chip (SoC) can include a central chiplet with a functionally safe shared memory through which other chiplets of the SoC communicate. The SoC can also include a cache memory accessible by the chiplets, a sensor input chiplet to receive sensor data from sensors and store the sensor data in the cache memory, a machine learning accelerator chiplet to calculate inferences using machine learning, and an autonomous drive chiplet to calculate autonomous driving algorithms.