Engineers have been designing em-bedded systems since Intel introduced the 4004 in 1971. So why is the 38th Design Automation Conference (DAC) placing a special emphasis on embedded systems in 2001? What's new is the mainstreaming of system-on-a-chip (SoC) and system-in-package (SiP) solutions, which enables the design of embedded systems on a few or even one die.
Embedding a complete system in an SoC or SiP, however, adds even more constraints to high-density chip integration. Embedded systems are programmable and may include RF and mixed-signal elements. These systems have to react to the external environment in real time and maintain very tight constraints in terms of power, cost, and size.
Classic embedded systems design gave engineers a broad range of choices in terms of hardware and software components and configuration. It also enabled them to independently assess and develop all elements of the design as subsystems before final integration. Compare this to a design where many components must be brought together on a single chip very early in the design process. Accordingly, issues such as design time, complexity of integration, and reliability would need to be taken care of much sooner in the design cycle—with severe penalties for mistakes and oversights.
By their nature, embedded systems require chip designers to deal with heterogeneous platforms. Unlike those who work with board-level embedded systems, chip designers have generally thought in homogeneous terms (e.g., a single processor with associated memory). But now, when creating an embedded SoC, designers must meld peripheral, DSP, communication, and processing IP from different sources. At the same time, it's essential for them to develop or evaluate OS and API software. This requires multidisciplinary skills and a clear vision of methodology. That's why DAC is refocusing on embedded systems.
At this year's conference, a number of sessions will reflect the dramatic changes in the embedded world as it converges with SoC-scale integration. For example, it may become necessary to re-assess operating systems designed for conventional board-level solutions because of the power, performance, and cost constraints of SoC/SiP integration. One critical difference is that the operating-system code must be compact enough to reside in the amount of real estate allotted to on-die memory. Plus, it has to manage the die's heterogeneous resources.
Methodology is crucial to successful embedded SoC/SiP systems as well. Despite object-oriented and top-down revolutions over the years, many embedded issues continue to be addressed ad hoc and bottom up. That's really the road to disaster. Designers need to convert to what's called a "scientific" design style to put everything together in a more formal and rigorous fashion.
Doing so will involve several new methods that the conference plans to explore. One of these is the development of abstractions: how to describe the problem and layer it into subproblems. Formalization is crucial in helping designers think about how to build correct systems from the start.
Other talks will cover subjects such as what the ultimate integration level should be. Also slated for discussion are the topics of platform-based design, the network-on-a-chip (how to communicate between heterogeneous components), and design-space exploration.
So if you're involved in embedded design in any way, you won't want to miss this year's DAC in Las Vegas, June 18-22. It's the best way to find out the latest in embedded developments and integration.