Overview

AI-enabled automotive systems increasingly rely on semiconductor systems-on-chip (SoCs) and intellectual property (IP) to execute neural-network workloads. Transferring a trained model to target hardware introduces additional considerations involving model conversion and quantization, hardware architecture, interface behavior, and the effects of systematic AI-related faults and random hardware faults.

This three-day certification training establishes the necessary artificial-intelligence and neural-network foundations before examining how convolutional neural networks (CNNs) are realized in semiconductor hardware. Participants study the roles of SoCs, integrated circuits (ICs), and IPs; potential failure modes and diagnostic measures across CNN processing paths and interfaces; and the coordinated application of ISO 26262:2018 and ISO/PAS 8800:2024 to semiconductor development.

The course also addresses AI safety requirements, safety cases and assurance arguments, verification and validation, and the supporting safety evidence needed for in-context and out-of-context semiconductor development. SRES instructors lead the training as senior industry practitioners with experience across automotive functional safety, semiconductor development, and AI-enabled systems. An optional exam to obtain the Semiconductor Artificial Intelligence Safety Professional (SC-AISP) certificate from SGS-TÜV Saar is available following the training.

Intended Audience

This course is intended for engineers, managers, and technical leaders involved in developing, integrating, evaluating, or assuring semiconductor products for AI-enabled automotive systems. It is particularly relevant for:

• Functional safety and AI safety engineers working on automotive SoCs, ICs, CNN accelerators, or reusable semiconductor IP
• SoC and ASIC architects, hardware designers, and embedded engineers developing AI inference platforms
• AI and machine-learning engineers responsible for optimizing, quantizing, converting, deploying, or validating models on target hardware
• Hardware safety, reliability, verification, validation, test, FMEDA, and fault-injection engineers
• Semiconductor product safety managers, technical leads, assessors, quality professionals, and compliance specialists working with ISO 26262 or ISO/PAS 8800
• Semiconductor manufacturers, IP suppliers, automotive suppliers, and OEM personnel responsible for safety requirements, safety documentation, or supplier-customer interfaces
• Professionals who wish to pursue the optional Semiconductor Artificial Intelligence Safety Professional (SC-AISP) certificate

Objectives

By the end of this course, participants will be able to:

• Explain foundational artificial-intelligence, machine-learning, and neural-network concepts, including deep neural networks and convolutional neural networks
• Describe the machine-learning development process and the steps involved in transferring a trained model to semiconductor hardware
• Understand model optimization, quantization, conversion, deployment, and testing on specialized neural-network hardware
• Explain the roles and architecture of SoCs, ICs, and integrated IP blocks in the realization of CNN-based functions
• Identify systematic AI-related faults and random hardware faults in CNN SoCs and evaluate their potential effects, propagation paths, detection mechanisms, and mitigations
• Assess safety considerations associated with sensor interfaces, memories, processing elements, accelerators, controllers, and communication IP
• Relate ISO 26262:2018 semiconductor safety activities to the AI-specific safety considerations addressed by ISO/PAS 8800:2024
• Understand the derivation of AI safety requirements, in-context and out-of-context development, assurance arguments, verification, validation, and supporting safety evidence for semiconductor products

Agenda

Below you will find an outline of the training course schedule.

DAY 1:

• Introduction to Artificial Intelligence
• Neural Networks: The Brain of AI
  • Introduction to Neural Networks
  • How Neural Networks Work
    • Simple Neural Networks
    • Multilayer Neural Networks
  • Deep Neural Networks Versus Convolutional Neural Networks
  • Convolutional Neural Networks (CNNs)
• Machine Learning and Model Transfer to Hardware
• Roles of Semiconductors (SoCs and IPs) in a CNN
  • Realization of a Model on CNN SoC Hardware
• Analysis of Potential Failure Modes in a CNN SoC
  • CNN Failure Modes and Their Potential Detection
  • Analysis Process According to ISO 26262

DAY 2:

• Analysis of Potential Failures in a CNN Use Case: ADAS System
  • System Architecture of ADS
  • System Information Flow of ADS
  • Generic ADAS System
  • Interface to the CNN: Analysis of Potential Failure Effects and Diagnostics
  • CNN SoC: Analysis of Potential Failure Effects and Diagnostics
    • Memory
    • Pooling
    • Filters
    • Perceptrons

DAY 3:

• Analysis of Potential Failures in a CNN Use Case: ADAS System
  • CNN SoC: Analysis of Potential Failure Effects and Diagnostics (Continued)
    • Memory Controller
    • Processors
    • Ethernet, PHY, and Peripheral Interfaces
    • MIPI Interfaces
• Semiconductor and IP Development According to ISO/PAS 8800
  • AI Safety Management
  • AI Safety Requirements Specification
  • AI Assurance Arguments
  • AI Verification
• Training Wrap-Up and Q&A

END OF DAY 3:

• Optional SC-AISP Certificate Exam