This post proposes abstraction with HSI/AII as an SDV design strategy to address market diversity.
Background
In recent years, the challenge has emerged in automotive development that the available AI algorithms and semiconductor devices differ from market to market due to the diversification of tariff policies and technology regulations in various countries and regions. These external factors have further increased the demand for flexibility in vehicle electro-architecture and software configuration.
Remedy
To address this issue, it is essential to standardize and clarify the Hardware Software Interface (HSI), which enables hardware abstraction, and the AI Interface (AII), which ensures implementation independence of AI modules, as a preliminary step to build a software-defined vehicle (SDV) architecture. Interface (AII), which ensures implementation independence of AI modules. This will enable rapid and flexible software adaptation to meet different market requirements.
Please note that I am not familiar with AII and may not be able to help you.
| AII | AI Data Interface | Interface for sensor data acquisition Interface for vehicle data |
| AI Inference API | API for inference engine Model input/output format Mid-tier APIs (AI vendors and SoCs) | |
| Control and Output Interface | Control-command interface Control-response interface | |
| Model management and update interface | OTA-Interface Model validation and logging interface | |
| Hardware Accelerator Interface | Drivers/APIs for interfacing with AI accelerators such as GPU/TPU/NPU | |
| HSI | MCU | Hardware Abstraction Layer (HAL) definition required for the target ECU |
| IC | Definition of IC specifications required for the target ECU |
