High-Level Structural Modeling

High-level design (i.e. microarchitecture design) is critical to the success of modern hardware. While microarchitecture has always had a significant effect on performance, modern designs also rely on microarchitecture design to manage power dissipation and thermal issues. Researchers are also proposing more and more high-level techniques to deal with emerging issues, such as inductive loading.

Even without the latest technologies, building the ideal microarchitecture is difficult. Designers must carefully consider all the tradeoffs to realize a well performing design. Processor designers often trade off features that only affect final performance by a few percent in the hopes that a small collection of these techniques will yield the 20-30% performance improvement expected from the microarchitecture in each new processor generate.

To properly make these difficult design tradeoffs designers need the ability to rapidly evaluate design criteria. Researchers too need the ability to rapidly model designs in order to evaluate novel ideas in a variety of contexts. Unfortunately, existing methodologies do not provide a practical way to rapidly build accurate models that

Our research shows that a concurrent-structural modeling methodology, with the specification language features that enable low-overhead reuse, a mechanism to gain reuse for difficult to modularize logic, the right execution semantics, and an optimizing compiler can provide a powerful system with these features. Our research into each of these areas is presented below.

The ideas in these pages are embodied in the Liberty Simulation Environment (LSE), a publicly available modeling tool that has been used both at Princeton University and other institutions to rapidly produce accurate models of hardware.

More information about these techniques can be found by following the links below: