Software Reviews

by Paul McLellan, VaST Systems Technology
Sunnyvale, CA, USA

Abstract :

The challenge of designing silicon systems has primarily been a problem of designing the silicon itself, and embedded software to run on the chip has been treated as an afterthought. The march of Moore’s law means that embedded processors continue to become cheaper and so more and more of a system is most economically implemented in software. In turn, this leads to explosive growth in software-rich chips.

A model of the chip must be built that is both accurate and with a performance high enough to allow software developers to use the model as a surrogate for the actual chip. If the model is not accurate enough, then it is not possible to derive useful performance data from the model, including whether the system will perform as expected. If the model is not fast enough then a duplicate implementation strategy will be necessary, developing the software as much as possible on a host platform and then porting it to the slow model for some level of validation.

New technology makes it possible to create a virtual platform in software, with accuracy and speed permitting software development to be done directly on the virtual platform, without needing either hardware boards or porting.

SOFTWARE-RICH CHIPS

Moore’s law states that the number of transistors on a chip doubles every 18 months. In addition, each of those transistors is cheaper. Looked at the other way round, Moore’s law states that the cost of implementing a given function in silicon halves every 18 months. Notably, the cost of implementing a processor halves every 18 months. In addition, processors continue to get faster and lower power, adding further to the attraction of implementing silicon functionality by using processors and embedded software.

It is easy to be misled that system companies, who ship the system that is sold to the eventual consumer, want to build chips for “competitive advantage.” In fact, system companies want to gain differentiation for their products in the most cost-effective way possible. Since chip design is very expensive and the schedule is very risky, designing a chip from the ground up to gain differentiation is a last resort. If possible they would prefer to use a standard or near-standard chip, and gain the differentiation through other means. Of course, there are always some systems where the differentiation comes from creative semiconductor design. The highest bandwidth routers or the highest performance graphics chips cannot be built around a standard microprocessor since still higher performance is available to a competitor who is prepared to invest in special high performance blocks. However, many systems are not like this: the required performance for a cell-phone or a 100 megabit/second Ethernet hub does not change with process technology. Instead, changes in process technology make different implementation choices possible. What is the optimum choice will depend on some combination of cost (mainly chip-size), speed and power. In most cases, this means use of processors whenever possible.