Hardware Acceleration for Neural Systems

为了真正理解大脑是如何工作的，我们需要模拟与人脑大小相似的神经网络（1000亿个神经元，100万亿个连接）。在标准计算机上模拟这样的网络是不可能的，因为它的规模太大了。目前有几个研究小组正在建造非常昂贵和专有的硬件来解决这个问题，但这些项目的成果将不会被其他研究人员使用。为了使神经科学取得真实的进展，需要让更多的研究人员参与进来。为了做到这一点，该项目将建立一个系统，从商业硬件（FPGA），将成本只有几万美元，它将使这种设计和软件是免费的。

To really understand how brains work, we need to simulate neural networks of a size similar to that of the human brain (100 billion neurons, 100 trillion connections). Simulating such a network on standard computers in not possible because of its sheer size. Several groups are currently building very expensive and proprietary hardware to solve this, but the output from these projects will not be accessible to other researchers. In order to make real progress in neuroscience, many more researchers need to be enabled to participate. To do this, the project will build a system from commercial hardware (FPGAs) that will cost only a few ten thousand dollars and it will make this design and software available for free.