How to Avoid the Traps and Pitfalls of SOC Design

How to Avoid the Traps and Pitfalls of SOC Design

Chances are pretty good that your current SOC design approach is making your job much harder than it needs to be. Start out on an easier path with this white paper.

A Short History of IC, ASIC, and SOC Design

The history of IC design is full of massive tectonic shifts. They've been occurring every 12 years or so since semiconductors arrived on the scene half a century ago. SOCs are ICs, so they're not immune to these shifts. It's been a little over 12 years since we first started to put processors on ASICs-which created the SOC. So we are about due for a tectonic shift.

The shift is caused by the exponential increases in SOC design complexity. In 1995, SOC designs usually involved one microprocessor core and fewer than a million gates. RTL design techniques-which became popular in the late 1980s-worked pretty well at that level of design complexity. Today's SOCs incorporate hundreds of millions of gates and multiple processor cores. As a result, design techniques that worked in 1995 have become cumbersome and clumsy, SOC design cycles are stretching out, and project schedules are slipping.

Similar situations have occurred before-at least three or four times. Early in the history of the IC, engineers designed chips and photomasks strictly by hand, using Xacto knives to cut polygons from rubylith. Then the industry graduated to early CAD tools that drew the polygons using a photoplotter. This single advance enabled the design of more complex chips. Then engineers started to drop entire transistors into their designs instead of drawing the individual polygons that comprised a transistor. This advance in design technique made even more complex IC designs possible. Soon engineers were designing with gates instead of transistors. By the late 1980s, the industry had left schematic design behind entirely and switched to RTL design and logic synthesis. Finally, IC designers started to adopt large blocks of logic. We call those blocks IP.

Each step up in design-abstraction level permitted the design of more complex ICs. Those advances allowed IC designers to keep pace with the IC-manufacturing advances that put more transistors on each new piece of silicon. In this way, IC design and manufacturing pretty much kept in step with each other for decades. Sometimes one would get a little ahead, usually manufacturing, but no serious problems arose because advances in design and manufacturing appeared in time to solve problems before seemingly insurmountable problems became truly insurmountable.

Endless Rise in Complexity and Performance Requirements

Today, as always, you must design chips almost as complex as theoretically possible to be competitive. If you don't, someone else will and the extra complexity will provide a competitive edge. That statement has remained true as the definition of "really complex" has evolved from hundreds of thousands of gates in the 1980s to hundreds of millions of gates today. The application drivers have also evolved. In the 1960s, the driver was military hardware. In the 1970s, it was mainframe computers. In the 1980s, it was personal computers. In the 1990s, it was cellular telephone handsets. Today, the applications driving the industry are wireless communications, multimedia, and all manner of consumer electronics.

As a result, there's an endless thirst for more computing horsepower to run the increasingly complex algorithms needed to handle digital media. And it's not just processing that you need to worry about. You also need to worry about moving large amounts of data onto and off of the chip and within the chip and everyone needs to worry about power dissipation.

Conventional approaches to system design no longer work in this new, more complex world. Design teams have had to abandon general-purpose processors and DSPs for media processing because they're just not energy efficient and they don't deliver the required performance unless clocked at multi-GHz rates. Fixed-function hardware is more efficient but often lacks the flexibility needed to help SOC designs span more than one narrow product niche, which is necessary to get the sales volume required to justify the design effort. Without such flexibility, you can't recover the development costs of these incredibly complex SOCs.