Why partners matter in today's semiconductor industry

I still remember when I entered the world’s biggest semiconductor company by value in 1980. This was an era where not invented here was the motto for many silicon vendors. And if they didn’t invent it, they made sure to convince everyone else that it wasn’t needed.

Semiconductor companies also had a habit of doing everything, from designing, manufacturing and packaging them to selling them directly. They even made the testers and the EDA software needed.

In short, silicon vendors were completely vertically integrated. Sometimes they went even further, making end products like transistor radios, toys, calculators, or home computers.

In this atmosphere, the not invented here syndrome was alive and well – and there was a feeling that companies could walk on water. Then some aggressive competition kicked in, and I started to discover that some of the things I was told were not really true. Some of the solutions that we offered were not received well by the industry; I vividly remember the case where one of our FAEs said his customer needed more RAM for a DSP. He was bluntly informed “we have put enough RAM on there for all conceivable applications”. It later emerged that the customer had a much greater vision of what a DSP could do, and yes, it really did need more RAM!

This sense of certainty pervaded everything. Some things were so far ahead of the market that they were discontinued even before they made it into production.

But one thing I noticed: when my company second sourced something, it never worked out. It seemed that buying companies was always the best solution rather than partnering with them!

Gradually I formed a healthy skepticism. If someone said it’s wonderful!, I always asked for my trusted customers’ opinions before I would use any superlatives to describe a new product. This skepticism would serve me very well when I started to deal with academics. Academics hate to be sold to, and marketing jargon or hype is a complete turn-off for them. They can also be mavericks: if you say “it’s the market leader and here are the figures to prove it” they will often look at the runners-up and see if they can champion them. They’re no fans of monopolies!

And then semiconductor IP came along. My company needed a CPU architecture for their mobile chipset and adopted CPU IP, something that represented quite a departure from their old ways.


Then later in my career, I finally moved on from the world of silicon vendors, and rather fortunately found myself in the world of IP.

At first this felt very strange. Even after 3 years, I have to say, it takes some getting used to. Not invented here is now an advantage. In fact, any SoC designers expect most of their IP not be invented in house! There’s IP buried inside IP inside other IP and built into a macro block of IP.

This necessitates a world of cooperation and collaboration. To make this stuff work it has to be interchangeable, interoperable, documented in a standard way, and modelled with industry-standard tools, for a precisely designed silicon process.

All of this became very clear when we started our MIPSfpga project. Releasing a real un-obfuscated MIPS microAptiv CPU for academic use, I quickly discovered that you cannot do this alone.

Firstly, our major licensees needed to know. Microchip Technologies used the MIPS microAptiv CPU in their PIC32 microcontroller family, and I needed their support before proceeding further.

Then there were the tools. We needed a tool chain, and open source GCC and OpenOCD came to the rescue there. Then there were the platforms. The world of FPGAs is fascinating, and depends not only on the right hardware, but excellent IDEs as well.

From the outset Xilinx were a delight to work with. They really know this business and they could see the potential in what we were doing.

Getting tools like Vivado IP Integrator to package the core as an IP block and then create the required blocks around it (memory, interrupts controllers, GPIO, etc.) required lots of support and debug. They helped us enormously.

For the guy used to everything happening in-house, it was a chastening experience to realise that our success was dependent on other companies.

We had a major issue with the JTAG interface for debug, and here it was Digilent who came to the rescue and became the critical problem-solvers that helped us release the final package!

So, to our wonderful partners on this project, I express my sincere thanks. Not only did you solve the problems and enable academics to use MIPSfpga, but you also taught an old dog some new tricks!

For more news and updates on MIPS, follow us on Twitter (@UniPgm, @ImaginationPR, @MIPSGuru, @MIPSdev), LinkedIn, Facebook and Google+.

Leave a Comment