DeePhi Technologies Ltd. Information

Founded in 2016, DeePhi Tech with recognized leadership in deep-learning acceleration, provides end-to-end solutions utilizing deep compression technology and deep-learning processor unit (DPU) platforms. Through the synergistical optimization of neural networks’ deep compression, compiling tool-chain, deep learning processing unit (DPU) design, FPGA development, and the co-design of algorithms, software, hardware and chip development, DeePhi is able to provide more efficient, convenient and economical inference platforms to all clients. DeePhi’s world-leading technology of compression and acceleration in deep-learning neural networks has received wide recognitions from world’s top academic conferences, which deeply influence the AI industry. Products from DeePhi, the Embedded Ends CNN Accelerators with Aristotle Architecture, the Cloud RNN/LSTM Accelerators with Descartes Architecture, and the 1st deep-learning development SDK in China - DNNDK™, work together as a system to free AI application in various devices from both embedded and server end. In addition, DeePhi Tingtao Series SoC is going to release in early 2018, which could achieve extremely high performance in deep-learning computing. 深鉴科技（DeePhi）成立于2016年，作为全球领导性的硬件加速方案和深度学习平台提供者，目前已具备世界顶尖的神经网络压缩编译技术、体系结构与指令集结构设计方法以及FPGA/芯片系统设计集成的研发能力，形成了以自主研发的深度压缩技术与深度学习处理器（DPU，Deep-Learning Processor Unit）为核心，从算法到软件到硬件的完整解决方案，让所有亟需解决高算力、低功耗需求的智能化设备提供商能够方便使用。 深鉴已发表的深度神经网络压缩与加速技术，获得了国内外顶尖学术会议的认可，在人工智能领域有着广泛而深入的影响力。自主研发的灵活、可扩展的DPU，包括面向CNN处理的亚里士多德架构（Aristotle），面向RNN处理的笛卡尔架构（Descartes），以及面向DPU的国内首款深度学习开发SDK——DNNDK™(Deep Neural Network Development Kit)，使得人工智能应用可以在嵌入式端和云端的不同设备中自由地实现。同时，为满足更多高算力需求，深鉴首款深度学习专用芯片“听涛”系列SoC将在2018年面世。