Many engineering managers see mask costs as one of the greatest barriers to silicon innovation—and if you go with the status quo, it’s easy to see the reasons why. However, I want to highlight a valid and, for many projects, more economically viable alter native.
First we do need to look at today’s perceived wisdom. It tells you to budget $600,000 for a design’s 32-mask set at 130nm, or $1 million for a similar set at 90nm. Y ou should also allow for at least one re-spin (research by Collet International says that 61% of designs need a re-spin). Combined, these factors make up anywhere between $1.2 and $2 million of the final NREs. Numbers like that don’t just worry startups and their investors; they can scare managers at the biggest integrated device manufacturers (IDMs) as well.
However, what if you could seriously reduce those costs by sharing a wafer with prototypes from other companies? Indeed, what if the cost of fabrication was based only on the real estate, the actual square millimetres, on this wafer occupying your project? And, then, what if you could drill down usage of this service so that your portion of the wafer only included those parts of a design’s circuitry that were new and unproven?
Well, you can do all of those things. Multi-project wafer (MPW) services offer just this flexibility, and are available from several foundries and specialist companies dedicated to providing the service through foundry partners. Figure 1 illustrates the principle.
The MPW concept has been around for more than 25 years and at least one MPW service provider has fabricated more than 50,000 designs. It’s a mature, proven, and successful, concept and is used by companies big and small. But it’s also true that MPWs are one of our industry’s best-kept secrets. Many project managers don’t know that they have the option. Among those who do, misconceptions are legion. I’ll give you an example.
Remember my comment that your outlay is based on the area of the wafer you occupy? Some say that the cost is split according to the number of users. That simply is not true. If you want to put only 40 test chips on a 300mm MPW, then that, based on their size, determines what you pay. Others feel that designs are restricted to predefined sizes. This is also untrue. Here are a few more ways in which MPW can make your project happen.
MPWs are available on both mature and leading-edge processes at major foundries. Independent service providers work with the likes of TSMC, IBM, and AMIS, and can help you build relationships with these companies for the move to high-volume production.
MPWs speed time-to-market. Turnaround times for a standard wafer batch are typically six to eight weeks, against 16 weeks for a production run. This doesn’t put a brake on your schedule. In fact, most likely, it will speed it up. Figure 2 demonstrates the economics and one of the key benefits: You can often get customer approval of samples produced using an MPW before you commit to the cost of a dedicated mask set and wafer run.
Services are flexible with quantity. Maybe you want just 40 test chips. Or maybe you need 1000 parts for qualification or early production. MPW service providers can match either of these requirements and plenty more in-between.
The MPW model can help you bring exciting innovations to market and control your budget. And this service isn’t only restricted to startups. Many of the largest names in the semiconductor industry regularly employ MPW services, too.
About a third of MPW designs are with IDMs—the companies you would think are the hardest to convince about MPW. Almost half are from companies employing 500 staff or more. And it’s a very attractive option, sometimes the only option, for many of silicon’s newest and smallest players. That is because, fundamentally, MPW services encourage all fresh-thinking engineers to take exciting risks with their designs with this knowledge: “Yes, you can afford to do it. Yes, there is another way.”