How Semiconductor IP Became Critical To SoC Design

Commercial IP has come a long way in the past 24 years.

popularity

By Mark Templeton
In 1991, I had the good fortune to be a member of the founding team of Artisan Components. We started the company believing that demand was about to appear for semiconductor intellectual property.

We had a few data points. We knew that before a company could start a new chip project, they first had to design and verify all kinds of generic building blocks – things like memories, I/O cells and standard cells. These building blocks were critical to the proper functioning of the final chip, but were hardly differentiating. And in 1991, these blocks weren’t commercially available for use with mainstream EDA environments.

We also knew that for a chip to have the best performance and density, the blocks had to be designed with an intimate knowledge of the design tools. Matching the connector locations on a standard cell library to the algorithms of the leading place and route tools led to dramatic density and route-time improvements. But this kind of tuning required months of experimentation as well as deep discussions with the tool developers – not something that most chip design teams could fit into their already tight schedules.

This was also about the time that pure-play foundries like TSMC and UMC were gaining acceptance. And once a design team decided to use a third-party foundry, the adoption of third-party IP was a logical step. In fact, design teams often viewed us as the process experts since our building blocks had usually seen silicon in advance of their own designs.

In the beginning, we found that we were competing against our customers’ own library development teams. There was a lot of NIH. And if you are only measuring a few parameters it is really difficult to decide which blocks are better. For example, the smallest standard cell layouts often resulted in a terrible place-and-route result when compared to slightly larger cells with deliberately spaced connectors and some porosity. This meant that we always had to have a better result than our customers could achieve internally – and we had to be able to defend our design decisions at a moments notice!

In parallel with our emphasis on silicon results, we also had to build an “IP factory.” We built a culture of quality, repeatability and schedule performance. We had to.

Once the business was going well, we were supporting 50 or 60 unique process technologies – most of which had several variants for different operating points. And of course, most of the processes that we were supporting were being refined and updated several times a year. That meant that we were constantly updating, modifying and releasing new versions of every product that we had in the field.

At one point we were releasing a unique IP library to our customer base about every 6 hours. That was a remarkable achievement and really speaks to the effectiveness of the technologies and flows that our engineering teams had developed. We had built the ultimate IP production machine.

Today, growing chip complexity forces the use of third-party IP. A complex system chip is a collection of significant subsystems – each of which requires a team of domain experts to specify, design, build and verify. Few companies can afford the domain experts necessary to cover all the bases. The solution is to use IP that has already been proven. In this way, every chip design team can have access to the skills of experts that they couldn’t possibly have as members of their own staff.

The success of Kilopass validates everything that I know about the value of third-party IP. Few companies have the time, patience and expertise required to create a rich family of non-volatile memories. Without IP companies like Kilopass, there would be no way to embed this kind of functionality on most system chips. The result would be poor design efficiency and higher system costs. This is clearly a case where semiconductor IP is advancing the industry.

Today’s chip design companies view IP suppliers as important partners that add differentiation and value to their offerings. And IP suppliers develop product roadmaps based on their industry knowledge combined with input from their best customers. This symbiotic relationship has made it possible for designers to keep up with the challenges posed by Moore’s Law and has enabled today’s most exciting products including smart phones, tablet computers and Internet-Of-Things devices.

I am proud of the pioneering role played by Artisan Components at the beginning of the IP era and am gratified to see companies like Kilopass and ARM carrying the torch into the future.

—Mark Templeton is president and CEO of Motius.