{"title":"Accelerating Integer Neural Networks On Low Cost DSPs","authors":"Thomas Behan, Zaiyi Liao, Lian Zhao, Chunting Yang","country":null,"institution":"","volume":24,"journal":"International Journal of Electrical and Computer Engineering","pagesStart":2711,"pagesEnd":2715,"ISSN":"1307-6892","URL":"https:\/\/publications.waset.org\/pdf\/2049","abstract":"In this paper, low end Digital Signal Processors (DSPs)\r\nare applied to accelerate integer neural networks. The use of DSPs\r\nto accelerate neural networks has been a topic of study for some\r\ntime, and has demonstrated significant performance improvements.\r\nRecently, work has been done on integer only neural networks, which\r\ngreatly reduces hardware requirements, and thus allows for cheaper\r\nhardware implementation. DSPs with Arithmetic Logic Units (ALUs)\r\nthat support floating or fixed point arithmetic are generally more\r\nexpensive than their integer only counterparts due to increased circuit\r\ncomplexity. However if the need for floating or fixed point math\r\noperation can be removed, then simpler, lower cost DSPs can be\r\nused. To achieve this, an integer only neural network is created in\r\nthis paper, which is then accelerated by using DSP instructions to\r\nimprove performance.","references":null,"publisher":"World Academy of Science, Engineering and Technology","index":"Open Science Index 24, 2008"}