清华大学类脑计算研究中心——于2014年9月创立， 是目前国内唯一一个进行全方位类脑智能研究的团队，涉及基础理论、类脑芯片、软件、系统和应用等多个层面。创立该中心的宗旨是突破类脑智能关键技术，发展通用人工智能。由于此项研究涉及信息、生物、物理、数学、材料、微电子等多个学科领域，在单一学科的框架下无法解决问题。由此，该中心由清华大学校内7家院系所联合而成，融合了脑科学、电子、微电子、计算机、自动化、材料以及精密仪器等学科，拟建设成一个具备多学科深度交叉融合能力的研究中心，从不同层面进行全方位的立体创新，发展以脑认知与信息科学为基础的类脑智能计算新范式，研究类脑计算系统新架构、新硬件、新软件和新算法，开发类脑智能技术及应用产品，推动类脑计算系统的发展。
当前，计算机技术面临两个重要的瓶颈：存储器和中央处理器分离“冯诺依曼”架构导致的存储墙效应造成能效低下，和引领半导体发展的摩尔定律预计在未来数年内失效。为应对上述挑战，借鉴人脑发展类脑计算技术已被国际半导体技术发展指南制定为重要的替代方案，欧盟，美国，英国，日本，韩国等国过都在大力推进脑科学或神经科学等相关领域的科学研究。2016年被誉为是类脑计算机元年，美、英、德相继推出了第一款类脑计算机，这是通往人工通用智能的关键基础。该领域目前仍处于起步阶段，尚未形成公认的技术方案。
有鉴于此，中心正致力于基于天机系列芯片的类脑系统的研发，于2015年11月成功的研制了国内首款超大规模的神经形态类脑计算天机芯片，同时支持脉冲神经网络和人工神经网络（如深度学习）。此芯片可进行大规模神经元网络的模拟，具有超高速、实时、低功耗等特点，特别适用于发展类脑智能技术。在此基础上，中心还开发了面向类脑芯片的工具链，提供友善的用户接口，降低应用的开发难度并提升效率。
类脑计算机系统：类脑计算机系统是按照生物神经网络采用神经形态器件构造的智能机器，采用微纳器件模拟生物神经元和突触的信息处理功能，采用大脑皮层神经网络结构作为基础体系结构，通过多传感器接收环境刺激以及和其他主体的交互获得和发展智能。与经典计算机系统相比，类脑计算机系统有三大特点：1）低功耗：类脑计算机系统过在体系结构上借鉴生物大脑而大幅度降低能耗，人脑功耗约20瓦，而据IBM测算，实时模拟人脑需要三百多台天河2号同时工作；2）高集成度和高容错：类脑计算机的基本元器件是模拟生物神经元和突触的神经形态器件，其特征尺寸与生物对应物相当或更小，不仅可比晶体管更小，而且部分器件出错不影响系统基本功能；3）经典计算机实现类脑功能主要靠复杂的人工编程和计算模型，类脑计算机主要采用通过环境刺激和交互训练实现感知认知等基础性智能，其效率更高，获得的智能也更适应复杂环境。
Center for Brain Inspired Computing Research (CBICR) ——was established in September 2014 in Tsinghua University. It is the first research center in China to perform the brain-inspired intelligence covering all the related research area, which includes neural functional / computational theory, neural coding / encoding, machine learning algorithms, operational systems, programming software, system integration, architecture design, chip processing, material science and nanotechnology. As the program involves multiple academic disciplines, it is impossible to perform the research under single faculty. Therefore, the center unites 35 affiliated faculties from seven departments in Tsinghua University, including brain science, electronic engineering, microelectronics, computer science, automation, material science and precision instrument, targeting the innovations on all related research areas. By deeply fusing the cerebrology and information technology, the center focuses on the realization of brain-inspired computing system. Our mission is to make the technology benefit the human society.
Currently, the computer technology faces two issues. One issue is the separation between the memory and center processing unit (CPU), leading to von Neumann bottleneck. The other issue is that Moore’s law is running out of gas and cannot guide the further development of semiconductor industry. The International Technology Roadmap for Semiconductors (ITRS), which coordinated the efforts of the semiconductor industry, has changed the emphasis toward systems and applications of device scaling [1]. The brain-inspired computing has become one of the focus. Moreover, a new International Roadmap for Devices and Systems (IRDS) was initiated to create the roadmap targeting on redesigning computer hardware and software in May 2016 [2]. Worldwide, the research programs in the fields of neuroscience, computing, and brain-related medicine have been greatly invested, including Human Brain Project (Europe), BRAIN Initiative (USA), Brain/MINDS (Japan), etc. Under the continuous efforts, the first generation of brain-inspired computers have been online in 2016 by utilizing TrueNorth, SpiNNaker and BrainScaleS’s chips, respectively, which is the milestone toward the artificial general intelligence. On the other hand, the research is still at the early stage and there is no well-acknowledged solution yet.
In view of this, the aim of CBICR center is to tackle the challenges in brain-inspire computing technologies. The first generation neuromorphic chip, named Tianjic, has been developed in Nov 2015, which can support both spiking neuron network (SNN) and artificial neuron network (ANN), including convolutional neural networks (CNNs), multilayer perceptrons (MLPs) and recurrent neural networks (RNNs). The Tianjic chips can implement large scale neural network with high speed and low power. Besides, a toolset, called NEUTRAMS (Neural network Transformation, Mapping and Simulation), is developed for better resource utilization.
当前，计算机技术面临两个重要的瓶颈：存储器和中央处理器分离“冯诺依曼”架构导致的存储墙效应造成能效低下，和引领半导体发展的摩尔定律预计在未来数年内失效。为应对上述挑战，借鉴人脑发展类脑计算技术已被国际半导体技术发展指南制定为重要的替代方案，欧盟，美国，英国，日本，韩国等国过都在大力推进脑科学或神经科学等相关领域的科学研究。2016年被誉为是类脑计算机元年，美、英、德相继推出了第一款类脑计算机，这是通往人工通用智能的关键基础。该领域目前仍处于起步阶段，尚未形成公认的技术方案。
有鉴于此，中心正致力于基于天机系列芯片的类脑系统的研发，于2015年11月成功的研制了国内首款超大规模的神经形态类脑计算天机芯片，同时支持脉冲神经网络和人工神经网络（如深度学习）。此芯片可进行大规模神经元网络的模拟，具有超高速、实时、低功耗等特点，特别适用于发展类脑智能技术。在此基础上，中心还开发了面向类脑芯片的工具链，提供友善的用户接口，降低应用的开发难度并提升效率。
类脑计算机系统：类脑计算机系统是按照生物神经网络采用神经形态器件构造的智能机器，采用微纳器件模拟生物神经元和突触的信息处理功能，采用大脑皮层神经网络结构作为基础体系结构，通过多传感器接收环境刺激以及和其他主体的交互获得和发展智能。与经典计算机系统相比，类脑计算机系统有三大特点：1）低功耗：类脑计算机系统过在体系结构上借鉴生物大脑而大幅度降低能耗，人脑功耗约20瓦，而据IBM测算，实时模拟人脑需要三百多台天河2号同时工作；2）高集成度和高容错：类脑计算机的基本元器件是模拟生物神经元和突触的神经形态器件，其特征尺寸与生物对应物相当或更小，不仅可比晶体管更小，而且部分器件出错不影响系统基本功能；3）经典计算机实现类脑功能主要靠复杂的人工编程和计算模型，类脑计算机主要采用通过环境刺激和交互训练实现感知认知等基础性智能，其效率更高，获得的智能也更适应复杂环境。
Center for Brain Inspired Computing Research (CBICR) ——was established in September 2014 in Tsinghua University. It is the first research center in China to perform the brain-inspired intelligence covering all the related research area, which includes neural functional / computational theory, neural coding / encoding, machine learning algorithms, operational systems, programming software, system integration, architecture design, chip processing, material science and nanotechnology. As the program involves multiple academic disciplines, it is impossible to perform the research under single faculty. Therefore, the center unites 35 affiliated faculties from seven departments in Tsinghua University, including brain science, electronic engineering, microelectronics, computer science, automation, material science and precision instrument, targeting the innovations on all related research areas. By deeply fusing the cerebrology and information technology, the center focuses on the realization of brain-inspired computing system. Our mission is to make the technology benefit the human society.
Currently, the computer technology faces two issues. One issue is the separation between the memory and center processing unit (CPU), leading to von Neumann bottleneck. The other issue is that Moore’s law is running out of gas and cannot guide the further development of semiconductor industry. The International Technology Roadmap for Semiconductors (ITRS), which coordinated the efforts of the semiconductor industry, has changed the emphasis toward systems and applications of device scaling [1]. The brain-inspired computing has become one of the focus. Moreover, a new International Roadmap for Devices and Systems (IRDS) was initiated to create the roadmap targeting on redesigning computer hardware and software in May 2016 [2]. Worldwide, the research programs in the fields of neuroscience, computing, and brain-related medicine have been greatly invested, including Human Brain Project (Europe), BRAIN Initiative (USA), Brain/MINDS (Japan), etc. Under the continuous efforts, the first generation of brain-inspired computers have been online in 2016 by utilizing TrueNorth, SpiNNaker and BrainScaleS’s chips, respectively, which is the milestone toward the artificial general intelligence. On the other hand, the research is still at the early stage and there is no well-acknowledged solution yet.
In view of this, the aim of CBICR center is to tackle the challenges in brain-inspire computing technologies. The first generation neuromorphic chip, named Tianjic, has been developed in Nov 2015, which can support both spiking neuron network (SNN) and artificial neuron network (ANN), including convolutional neural networks (CNNs), multilayer perceptrons (MLPs) and recurrent neural networks (RNNs). The Tianjic chips can implement large scale neural network with high speed and low power. Besides, a toolset, called NEUTRAMS (Neural network Transformation, Mapping and Simulation), is developed for better resource utilization.
