A Neural Network Engine for Resource Constrained Embedded Systems
| العنوان: | A Neural Network Engine for Resource Constrained Embedded Systems |
|---|---|
| المؤلفون: | Zuzana Jelcicova, Evangelia Kasapaki, Adrian Mardari, Oskar Andersson, Jens Sparsø |
| المصدر: | Jelcicova, Z, Mardari, A, Andersson, O, Kasapaki, E & Sparsø, J 2020, A Neural Network Engine for Resource Constrained Embedded Systems . in Proceedings of 54 th Asilomar Conference on Signals, Systems, and Computers . IEEE, pp. 125-131, 54 th Asilomar Conference on Signals, Systems, and Computers, Pacific Grove, California, United States, 01/11/2020 . https://doi.org/10.1109/IEEECONF51394.2020.9443426 ACSSC |
| بيانات النشر: | IEEE, 2020. |
| سنة النشر: | 2020 |
| مصطلحات موضوعية: | Digital signal processor, Artificial neural network, business.industry, Computer science, Quantization (signal processing), 020208 electrical & electronic engineering, 02 engineering and technology, Reduction (complexity), Memory management, SDG 3 - Good Health and Well-being, Keyword spotting, Embedded system, Datapath, 0202 electrical engineering, electronic engineering, information engineering, Benchmark (computing), 020201 artificial intelligence & image processing, business |
| الوصف: | This paper introduces a dedicated neural network engine developed for resource constrained embedded devices such as hearing aids. It implements a novel dynamic two-step scaling technique for quantizing the activations in order to minimize word size and thereby memory traffic. This technique requires neither computing a scaling factor during training nor expensive hardware for on-the-fly quantization. Memory traffic is further reduced by using a 12-element vectorized multiply-accumulate datapath that supports data-reuse. Using a keyword spotting neural network as benchmark, performance of the neural network engine is compared with an implementation on a typical audio digital signal processor used by Demant in some of its hearing instruments. In general, the neural network engine offers small area as well as low power. It outperforms the digital signal processor and results in significant reduction of, among others, power (5×), memory accesses (5.5×), and memory requirements (3×). Furthermore, the two-step scaling ensures that the engine always executes in a deterministic number of clock cycles for a given neural network. |
| وصف الملف: | application/pdf |
| اللغة: | English |
| DOI: | 10.1109/IEEECONF51394.2020.9443426 |
| URL الوصول: | https://explore.openaire.eu/search/publication?articleId=doi_dedup___::bc47cefe9d9297c0c7d743956ff16310 https://orbit.dtu.dk/en/publications/4cea061a-a5ec-4426-8567-2cb2ca6ac494 |
| Rights: | OPEN |
| رقم الانضمام: | edsair.doi.dedup.....bc47cefe9d9297c0c7d743956ff16310 |
| قاعدة البيانات: | OpenAIRE |
| DOI: | 10.1109/IEEECONF51394.2020.9443426 |
|---|