The CPU market has a lot of players: AMD, Analog Devices, Atmel, Broadcom, Cirrus Logic, Freescale, Fujitsu, Intel, IBM, Infineon, MIPS, NEC, Nintendo, NXP Semiconductors, OKI, Samsung, Sharp, STMicroelectronics, Sun, Texas Instruments, Toshiba, VIA. And I'm sure I missed a bunch. For more information on what they make, just type the name and the word microprocessor into Google (ie. NEC microprocessor). Your router has a CPU, your VOIP handset has a CPU, your cell phone (if it's recent) has a CPU, a lot of the newer TV's have CPU's, Blue-ray players have CPU's, and if you have a recent car it'll have multiple CPU's.
But there are few companies that make desktop Windows-compatible CPU's. I think it's mostly a matter of infrastructure. To be competitive you need to be on a cutting-edge process and you need a large experienced design team, a portfolio of patents and a market advantage over everyone else. You need the cutting-edge process technology to keep your design small (and cheap to manufacture) and for the power and performance advantages of each of the generations of process technology. Access to cutting-edge process technology requires you either: build a fab, or pay a lot to another company to fab it for you. Fab's cost about $3.5 billion dollars each. Access to cutting-edge process technology from semiconductor foundries is difficult to get (you are competing with existing customers - like, say, nVidia) and expensive.
Then there's the design of a modern deskptop CPU. If you are targetting a niche market, then designs can mostly be completed by "synthesis" in which a bank of computers running synthesis tools (google it for more info), basically design the CPU for you based on code that looks a lot like C code. So you get a couple of hotshot architects to code up a CPU for you, put it into a synthesis tool and then you have your design. This strategy works well if area, power and performance are not critical... but usually they are. And VLSI synthesis packages aren't cheap.
But, synthesis tools still have a long way to go to catch up with human engineers and so if you want a small, high performance design that uses as little power as possible, you need people to do it... specifically, you want trained and experienced VLSI electrical engineers. And these engineers are not cheap; median salaries for 2007 were around $108k (
http://www.eetimes.com/news/la...l?articleID=202601561). And you need a lot of them. You need architects (figure out the design, and performance trade-offs), logic designers (take the architecture and turn it into microarchitectural code), circuit designers (take the microarchitectural code and turn it into circuitry), layout technicians (take circuitry and turn it into a mask for manufacturing). But these designs are also really complex, and, like any large project, there will be bugs. So you need all lot of validation engineers to find bugs and to check it against the existing software, and then you'll need debug engineers (find bugs and fix them). You'll need people monitoring the design for electrical robustness, and manufacturing issues. In short, you need a lot of people. About 75 engineers working for 18 months is about the minimum for a modern design, but I've heard of designs that are above 500 engineers working for 3 years... just on the design. The IBM Power team quoted even larger numbers than these back at ISSCC, so big designs are even more.
So then you have your $3.5 billion dollar fab, and your 200 engineer, 3 year long design, and you go to sell it... how many do you have to sell to pay off your investment? So you need to keep your CPU prices very competitive, but you need to pay off your investment to be a profitable company. And what if you missed a bug or two - when there's a billion transistors, it's not too hard - so then there's the possibility of recalls. Or there's a reliability issue (such as what just tanked nVidia's stock last week). Or the design tools let you down and your estimates of how great a product you are going to have don't match up with the reality of real silicon and it's either slower or hotter than your design estimates. And even if things go great, you have to market your design, and work with OEM's to integrate it and validate it in their products. And you need to be working on the next design in parallel. And you need to validate that it works in a complete system.
If you don't opt for the fab and want to use a fabless design and you want to rely heavily on synthesis - both of these to keep costs down - then you will find that it is extremely difficult to have a competitive product. Synthesis does ok for some things, but for things like arithmetic circuitry, caches and registers, you need a human touch to get something that works well. So, if you opt for the really inexpensive investment route, then you will likely end up with a chip that is large, hot and has lousy performance... and if not all three, then likely at least two out of the three (large and hot, but fast, for example). If it's not bleeding edge, or if it doesn't have some form of advantage over the competition, it's not going to sell well. And the competition is a moving target - your design can't just be better than anything available right now, when you start the design process, you need to be better than what will be out there in 3-5 years (when you finish).
And, as mentioned, then you run into the issue of patents, trade secrets, trademarks. Semiconductor companies have invested huge amounts on money into the way they design the architecture and microarchitecture, design circuitry, and manufacture and package designs. If they invent something new and innovative, they patent it and this means that you'll need to innovate a different way to do it that doesn't copy their patent. If they have a method of doing something that don't patent, it's still covered under "trade secrets". If you don't want to get into the chipset design business (another large design team), you'll need to license bus protocols. And current names are trademarks, and so you'll need to get customers to accept your new name. Would you buy a laptop with a CPU from a company you've never heard of?
Lastly, you need to test your design. CPU's are big and complex and as much as engineers would like them to, not all of those transistors will get drawn correctly during manufacturing. So you need to test the designs and this requires a team of engineers to write and optimize tests, and then to get them working and figure out how to improve the "manufacturability" of the product. Testers are expensive - they are so expensive and are such speciality items that I couldn't find any web pages about them. But testers like the Agilgent Advantest series, and the Schlumberger 9000-series are about the size of a large car and cost a lot more.
CPU development is a difficult and complex process which has a lot of risks. To succeed, you need to be a big company with history and experience in these types of projects to succeed, and you need deep pockets to survive the inevitable issues. There have been startups (Transmeta, for example) that have tried, but even with a lot of venture capital it is very difficult.
All that said, I personally believe that the CPU market is extremely competitive - whether it's desktop CPU's or more embedded CPU's. I think it's actually more competitive than it has been in a while, and that it will continue to get more competitive as we move into the next decade. On the x86 desktop/laptop CPU market, prices continue to drop from generation to generation. You can buy an EEE PC laptop for under $400 that runs XP. Laptops used to run over $1000 several years ago, and now they are much less. And I would argue that as we move into an era of "ubiquitous computing" - in which everyone carries around a "computer" (ie. like a Blackberry or iPhone) and has perpetual access to the internet - the role of the desktop PC becomes diminished. Which is more exciting to consumers, a 3G iPhone, or a new Dell PC design with a new CPU in it? And that microprocessor in the 3G iPhone is made by Samsung.
I can't predict the future... when I try to, I get it wrong. But as we move to carrying around computers in our pocket, and we run mobile OS's like Android and do more and more functions on the internet using browsers and software that are not dependent on a specific CPU architecture, and we have media centers on our TV's and gaming consoles, and our latest desktop/laptop computer is "good enough for Excel and doing my taxes", I think the market will become even more competitive as existing companies have to innovate and new companies appear with new and creative ideas.
It should be noted that I am not speaking as a representative of my company, and that my opinions are my own.
Facebook
Twitter