Hardwares for AI and ML( Brain Computing)/ Hardwares for computational neuroscience

The Most ununderstood human’s organ is the Brain. Neural networks were developed to mimic the salient features of biological brains, such as their ability to recognize patterns and detect motions in the presence of noises. As an algorithm, a neural network needs to be executed on certain hardware platforms before it can be deployed in various applications. There are, in general, three types of hardware that neural network algorithms can be deployed: general-purpose processors, ?eld-programmable gate arrays (FPGAs), and application-speci?c integrated circuits (ASICs). A comparison of these three hardware platforms is made in terms of energy e?ciency and ?exibility. 

Most scientists and researchers in the ?eld of arti?cial intelligence are most familiar with this type of hardware platform due to their wide accessibility. The drawback of a general-purpose processor, however, is its low energy e?ciency. The low energy e?ciency is apparently a problem for portable devices that are powered by batteries .In order to achieve higher energy e?ciency, an ASIC solution is often adopted. An ASIC is a customized chip that is designed to serve for a speci?c type of application, which, in our case, is neuromorphic computing. Since the architecture of an ASIC can be speci?cally optimized for the target application, an ASIC typically has the highest performance as well as the lowest energy consumption. The drawback of an ASIC is that many logic and arithmetic operations are hard-wired and it typically only implements a ?xed function (although some of them do provide certain levels of re-con?gurability).