Abstract
Since the NPU for smartphones was released in 2017, various hardware architectures for on-device AI on smartphones have been proposed and implemented. The AI performance of a smartphone varies depending on software development kit (SDK) and application programming interface (API). However, the software development environment of smartphones, such as SDK and API, is only known to a group of related experts. Still, it is not popular among researchers in academia. This chapter will introduce Google’s Android Neural Network Application Programming Interface (NNAPI) and a representative acceleration API for smartphone AI. NNAPI provides convenience in the Android platform, but vendors use their SDKs to run their hardware with optimal performance. This chapter will also introduce the various functions and advantages of Qualcomm’s SDK, one of the representative vendor SDKs. Machine learning benchmark Apps are trying to evaluate and compare the AI performance of vendors’ smartphones fairly. However, the score and ranking may vary depending on their policy, such as precision, SDK/API, machine learning models, etc. Users and developers should know this story to evaluate NPU performance objectively; this chapter explains the issue of SDKs of various vendors.
| Original language | English |
|---|---|
| Title of host publication | Artificial Intelligence and Hardware Accelerators |
| Publisher | Springer International Publishing |
| Pages | 151-165 |
| Number of pages | 15 |
| ISBN (Electronic) | 9783031221705 |
| ISBN (Print) | 9783031221699 |
| DOIs | |
| Publication status | Published - 2023 Jan 1 |
Bibliographical note
Publisher Copyright:© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.
All Science Journal Classification (ASJC) codes
- General Engineering
- General Computer Science