Many-core architecture for NTC: Energy efficiency from the ground up

The high energy efficiency of NTC enables multicore architectures with unprecedented levels of integration, such as multicores that include 1000 sizable cores and substantial memory on the die. However, to construct such a chip, we need to fundamentally rethink the whole compute stack from the ground up for energy efficiency. First of all, we need techniques that minimize and tolerate process variation. It is also important to conceive highly-efficient voltage regulation, so that each region of the chip can operate at the most efficient voltage and frequency point. At the architecture level, we want simple cores organized in a hierarchy of clusters. Moreover, techniques to reduce the leakage power of on-chip memories are also needed, as well as dynamic voltage guard-band reduction in variation-afflicted onchip networks. It is also crucial to develop techniques to minimize data movement, which is a major source of energy waste. Among the techniques proposed are automatically managing the data in the cache hierarchy, processing in near-memory compute engines, and efficient fine-grained synchronization. Finally, we need core-assignment algorithms that are both effective and simple to implement. In this chapter, we describe these issues.