ASIC option may not pay off as expected

ASIC option may not pay off as expected
By Kevin Klein, EE Times
December 11, 2003 (6:58 p.m. EST)
URL: http://www.eetimes.com/story/OEG20031211S0044

When asked why he wanted to climb Mount Everest after the failure of the first serious ascent attempt ever, George Leigh Mallory replied, "Because it's there." When I hear from customers that they've decided to have a go at doing an ASIC, I generally wonder what their motivation is. While for some it is the right answer to their needs, for many it is something they are doing because they can. Today, with the many varieties of 32-bit devices now available, system designers need to ask themselves some basic questions before embarking down the ASIC path.

There is a certain attraction to being able to get just what you want. But the reality is that the costs and hassles often negate those benefits. Even customers with lots of experience designing many high-volume ASICs are realizing that it just may not pay off in the long run.

There are really only two good reasons for doing a custom system-on-chip ASIC device for your product design. The f irst is that some technical issue in your system simply cannot be solved any other way at a reasonable cost. Perhaps it's a performance issue, where a function must be done in hardware instead of software. Or maybe a custom protocol or hardware interface must be used. Either way, it's something that just cannot be done any other way.

The second reason is that you have the staggeringly high volumes that justify the costs. In such a case, the per-unit price of an ASIC will often be lower than that of an MCU. So if you have a product or product family that will have a run into the millions of units, then it may make sense to invest in developing a custom ASIC. But even then it may not be a good idea.

Why go 32-bit?
Today's 32-bit processors have an impressive array of features available at a reasonable price. Not only do they have high-performance CPUs but also lots of on-chip memory, high-performance off-chip memory interfaces, multiple communications interfaces, on-chip timing and A /D conversion. So what are some of the benefits of an off-the-shelf MCU approach compared with an ASIC?

First let's look at cost, since that's often a primary motivator. Many system designers look at an off-the-shelf 32-bit microcontroller and note that it has features that aren't required in that system. They believe they can create a "stripped down" device that is just what they need, at a lower cost.

But in advanced process technologies, the cost of putting additional functionality on the MCU (logic gates) is coming down faster that the cost of getting the signals off the chip (pads). Unless your system architecture allows you to get by with not only a fraction of the logic functionality but also a fraction of the signal counts, then don't waste your time trying to develop an ASIC. By the time you cover the nonrecurring-engineering costs, mask costs and so on, you'll have more than eaten up whatever savings you hoped to realize.

Another huge benefit of off-the-shelf 32-bit processo rs is their flexibility. The same flexibility that makes them look like nonoptimum solutions may save the day if the requirements of the end product suddenly change, or if a bug is found in your custom-ASIC design. While you are off doing another costly ASIC spin, your competitor is selling product.

One concern about using an off-the-shelf processor is the fact that they are nonexclusive, meaning that your competitors can use the part as well. Since that may erode some of your product's differentiation, it is a legitimate concern. But the flip side is the benefits that come from the use of these products by many companies, competitors and noncompetitors alike.

One advantage is supply. Most off-the-shelf solutions are available the day you start your design, allowing you to get a running start. Many customers use them, so they will be available when you need them, through product ramps, and not sitting in your warehouse through industry downturns. And since they are not dependent on a single c ustomer to stay viable, the products will be available for years to come.

Also, the fact that others are using these products provides as much of a chance for you to learn from them as for them to learn from you. Since these MCUs have to be attractive to a broad base of customers, they are generally packed with the features and interfaces that best reflect the trends of the industry, so you benefit from the experience and knowledge of a wide base of users, as well as the design and marketing expertise of the supplier.

But the most important issue of all may be opportunity cost. While you have your key designers tied up with an ASIC design instead of system design, you could be getting this product, or maybe even the next product, out the door. If you are so good at designing silicon products, why are you wasting your time doing systems? And if you're really good at system design, why not stick to it?

Remember Occam's razor: The simplest solution is the best. So unless your product rea lly needs it, resist the temptation to "roll your own" solution, then grab a 32-bit MCU and get your product out the door.

Kevin Klein is with Motorola Inc.'s 32-Bit Embedded Controller Division.

See related chart