17-03-2017, 09:10 AM
Engineering is a compensation study. In computer engineering, compensation has traditionally been between performance, measured in instructions per second, and price. Due to manufacturing technology, the price is closely related to chip size and number of transistors. With the advent of embedded systems, new compensation has become the focus of design. This new compensation is between performance and power or power consumption. The computational requirements of early embedded systems were generally more modest, and therefore the balance between performance and power tends to be weighted toward power. "High performance" and "energy efficiency" were generally opposite concepts.
However, new classes of embedded applications are emerging that not only have significant energy limitations, but also require considerable computational resources. Devices such as space rovers, cell phones, automotive control systems and portable consumer electronics require or can benefit from high-performance processors. Future generations of such devices should continue this trend. Processors for these devices must be able to deliver high performance with low power dissipation. In addition, these devices show great fluctuations in their performance requirements. Often, a device will have very low performance demands for most of its operation, but it will experience periodic or asynchronous "spikes" when high performance is required to meet a deadline or handle some interrupt event. These devices not only require a fundamental improvement in performance power compensation.
MORPH
These objectives and motivations were identified during the MORPH project, part of the Power Computing / Communication (PACC) initiative. In addition to exploring several mechanisms to fundamentally improve performance, the MORPH project gave birth to the idea of "gearshift" as an analogy to runtime reconfiguration. Realizing that real-world applications vary their performance requirements dramatically over time, one of the project's main goals was to design microarchitectures that could be tuned to provide the minimum required performance at the lowest energy cost.
The MORPH project explored a number of microarchitecture techniques to achieve this goal, such as caching hierarchies and exploiting the inactivity bit-gap. One technique, the multi-cluster architectures, is the direct predecessor of this work. In addition to changes in microarchitecture, MORPH also conducted a survey of embedded realistic applications that may be power limited. In addition, we explored the design implications of a runtime system with energy consumption.