Moore's Law Is Collapsing, and the Cambrian Explosion of Computer Architecture Has Begun.

In the past, semiconductor integration density doubled every 18 months, following Moore's Law. Leveraging this integration density, computer performance improved exponentially, and now the performance of general-purpose computers used in laptops, smartphones, and tablets has reached a level that is satisfactory for practical use. On the other hand, the end of improvements in semiconductor integration density is now in sight. Without a breakthrough, the "post-Moore era," in which integration density will not improve at all, will arrive in a few years. So, should the progress of computers stop in this era? Not at all. Due to the development of deep learning and big data processing technologies, there is a growing need for computers with extremely high processing power for specific tasks such as autonomous driving, drones, robots, factory automation, and interactive processing. The post-Moore era will not be one where general-purpose computers do everything as they have in the past. Instead, it will be the age of Domain Specific Architecture (DSA), which performs processing limited to specific fields with orders-of-magnitude higher capability and low energy consumption.

DSAs only need to achieve performance in specific domains, and with the development of cloud computing, even specialized and expensive systems can be commercially successful by being shared among many users. For this reason, ideas that were previously only used at the research level, such as systolic arrays, logic-in-memory, and analog neuro, are now being implemented one after another in actual chips. Giant cloud companies like Google and Amazon are developing their own new computers for deep learning and using them in the cloud, while in the edge domain, ambitious entrepreneurs are creating chips with new ideas and rapidly bringing them to market. Seventy years after the birth of the stored-program computer, we are finally about to enter an era of a Cambrian explosion in computer architecture diversity. The Amano Laboratory is pioneering this era with an approach of massively parallel processing and reconfigurable hardware. Figure 1 shows the block diagram and a chip photo of a new architecture we developed called CMA (Cool Mega Array). The competition is fierce, but the target fields are also expanding, and everyone has a chance to succeed.