Spatial computation is a technique for improving performance by optimizing wires (reducing communication time) at the expense of computational units. The idea is that communication, and not computation, is now a bottleneck for improving performance. A very narrow area (application-specific hardware) is used in this study. The major contribution of this paper is a compiler combined with other tools that produces application-specific hardware directly from standard C code. The hardware produced is a dataflow machine that consumes much less power than conventional hardware. This will be very useful for autonomous computing.

The authors have combined ideas from many sources (the bibliography has over 100 references) to create a tool that works without changes to the C language. The tool does produce very power-efficient hardware, but at the moment the hardware produced (in most cases) does not perform as well as the 600 MHz machine used for comparisons. The authors present some possible reasons why the computational speed is slower (since it is intuitive that the application-specific hardware should be faster).

Many acronyms and terms are used without definition, so the reader needs to be familiar with the area, or be ready to access several of the references. There is an excellent section on the various performance aspects of this new tool, and the hardware it produces. Additionally, there is an excellent future research section.