智能驾驶人工智能芯片的计算性能测试研究

摘要：

智能驾驶汽车发展迅速，对汽车人工智能（Artificial Intelligence，AI）芯片的计算性能要求持续提升，也引发了行业对于验证芯片计算性能是否满足实际应用需求的关注。然而，目前现有研究多为通用性计算性能测评方法，针对汽车领域应用的AI芯片计算性能有待研究。本文基于这一行业问题，对测试模型进行调研，提出了一套面向智能驾驶应用的AI芯片计算性能测试方法，并选取典型模型对实际产品进行测试，对比分析不同芯片各模型下的测试结果。本文研究内容对于汽车行业进行AI芯片选型也具备重要的参考价值。

Abstract:

The rapid development of intelligent driving vehicles has continued to raise the computational performance demands for automotive artificial intelligence (AI) chips, and has triggered the industry’s concern about verifying whether the computing performance of chips meets the needs of practical applications. However, most existing researches focus on general computing performance measurement methods, and the computing performance of AI chips for automotive applications still needs to be researched. Based on this industrial problem, this paper conducts survey on test models, proposes a set of testing methods of AI chip computing performance for intelligent driving applications, and selects typical models to test different products, and analyzes the test results of different chips under each model. The research has important reference value for AI chip selection in the automotive industry.

基金项目：本文受市场监管总局技术保障专项“汽车芯片安全可靠性认证技术及通用审查方法研究”(项目编号：2023YJ38)资助。

作者简介：赵瑞，硕士研究生，主要从事汽车芯片检测技术研发，牵头及参与制定10余项汽车芯片行业标准，参与国家重点研发计划、省部级重大科研计划等课题。吴海文，通信作者，博士研究生/博士后，正高级工程师，从事合格评定政策与技术研究。