DeePhi Tech was founded in 2016 by reseachers from Tsinghua and Stanford universities. DeePhi offers a data compression tool and Deep neural network compiler for efficiently mapping neural network into high performance digital processor unit instructions.
DeePhi's Aristotle architecture is designed to compute convolutional neural networks (CNNs). While currently used for video and image recognition tasks, the architecture is flexible and scalable for both servers and portable devices. The Descartes architecture is designed to implement compressed Recurrent Neural Networks (RNNs) and is suitable for FPGA implementation of instantaneous speech recognition, neural language processing, and other recognition tasks.
DeePhi announced a Series A round of financing in May 2017 said to be worth tens of millions of dollars. Investors included FPGA vendor Xilinx Inc. and SoC developer MediaTek Inc. as well as Tsinghua Holdings Corp. Ltd., Sigma Square Capital, and existing investors GSR Ventures and Banyan Capital.
According to the Korea Herald Samsung invested a significant amount in August 2017 and its interest may be to do with having a neural processing unit (NPU) for inclusion on an upcoming generation of Exynos application processor. Such an NPU as a coprocessor core could be used in a smartphone to provide speech recognition, language translation and other natural language processing tasks.
The same report said that mobile carrier company SK Telecom had also offered to invest in DeePhi but had been turned down. However, the report also said that Samsung's investment may have been driven by political considerations and not by the excellence of the company's technology.
The investment meets a goal of Samsung to increase overseas investment. One year ago Samsung joined with Bosch Ventures to lead a $30 million investment round in Graphcore Ltd. (see Graphcore gets big backing for machine learning ).
Related links and articles: