‹header›
‹date/time›
Click to edit Master text styles
Second level
Third level
Fourth level
Fifth level
‹footer›
‹#›
In 1965, Gordon Moore, then research director for Fairchild Semiconductor was asked to comment about the future of this new gadget called integrated circuits, invented in 1959 by Jack Kilby and Robert Noyce.  They wanted him to comment on whether or not this new technology has a future.  As any good research director will do, he sit down and look to the history as a predictor of the future. <click> He noticed that the first 8 transistor chip appears in 1962,<click>.  The first 16 transistor chip appears in 1963, <click> the first 32 transistor chip appears in 1964.  Staring at one of his own chips in 1965 which has 64 transistors, Moore boldly predicted that number of transistors on an integrated circuit will double every year.
So that’s the prophecy known as Moore’s Law.
In 1968, Moore and Noyce left Fairchild to found this tiny startup company called Intel. <click>
Their first processor chip, 4004, appears on the market in 1971, contain 2,250 transistors. 
The next year see 8008, which contains 2500 transistors. 
8080, which appears in 1974, contain 5000 transistors.
In 1975,  acknowledging the increasing difficulty in chip manufacturing, and seeing his prediction coming up short, Moore revised his prediction to doubling every two years.
And the rest is history.
Year after year, the transistor count on an integrated circuit, well, the transistor count of Intel microprocessor products anyway, followed remarkably close to the Moore’s Curve.  And I am sure, not the least due to the fact that he is the chairman and CEO of Intel, afterall.
The Moore’s Law, of course, is not a true law in the sense of laws of physics or other laws of science.  It is just a self-fullfilling prophecy.  It is the highest form of technology prediction, where every researcher, every engineer believes that if he or she solve their portion, everything around will magically be solved by others and the whole thing will work together at the end.
Anyway.  So what does it looks like from here.
Semiconductor Industry Association is a consortium consists of all the big players up and down the semiconductor food chain: IBM, HP, Intel, Motorola and international players like Sony, Philips, TSMC.   They put out an “International Technology Roadmap for Semiconductors”.  In which they point out the technology trends based on Moore law, market force, and technology challenges.
Among other thing, it predicts that by the year 2014, a high-end IC will have between 5 to 12 billion transistors, compare to today’s Pentium IV at 42 million transistors, 4000 to 9000 pins, 1 order of magnitude more than Pentium IV,  30nm feature size, compare to state of the industry at 180nm today.  On-chip local clock at 15 GHz.  Vdd range from 0.3V to 0.6V, and power dissipation at a whopping 200W.
5 billion transistors, that’s a lot of transistors…
Just for fun, I went on the web and look to compare the rate of improvement for human to the rate of improvement for integrated circuits.  Since human brain cell hasn’t shrunk for millions of years,  the cranial size is a good indication of the computing power, if the quality of the brain cell is assumed to be constant. 
“Lucy”, widely considered the first “human”, has a cranial size of 380 cc and live approximately 4 million years ago.  <click> I then plot the cranial size of  some of the other human species against the time they live, all the way to ourselves, homo sapiens, <click> with an average cranial size of 1400 cc.  I obviously have to plot it on a linear scale  because otherwise it will be pretty flat. 
If you work out the math, human cranial size double historically at a rate of once every 2.08 million years. <click>  So to solve a design problem that is doubled in size due to Moore’s law, the theory of evolution says that the only thing we have to do is to wait 2.08 million years, there will then be an Intel designer with head twice as big and with sufficient cranial power to effectively design a chip with twice the transistors.
Fortunately, the quality of human brain cell does also improve rapidly.  Through the use of collective knowledge base and CAD tools and methodologies, it shows up as an exponential human productivity growth as well.
I grep this graph straight out of the roadmap.  It shows that while the chip complexity increased historically at 58% per year, designers productivity also grew at a rate of 21% per year.  There remains, however, the very well known, much publicized issue of productivity gap.  As chip complexity grows, more and more designers are required and each to design less and less transistors.  At some point in the near future, the gap will grow so large that throwing more designers into the mix is not going to help much.
So what to do?  There will probably be some elements of giving up, as the die size will probably stay constant or shrink a little while the wafer size increases.  There will also be more on chip memory.  Other than the “easy” solution of  hiring more designers, building smaller die, adding more memory to the chip, improvement in design science can help in the form of better CAD tools and methodology, and by raising the level of abstraction.
There are ample precedence's for the paradigm shift.  The first integrated circuit of the 60’s are built by drawing mask layers,<click> by the 70’s, the designers design with CMOS transistor and never have to think about mask ever again.  <click> The transistors are clustered and abstracted into gates in the 80’s.<click>  In the 90’s RTL design gain acceptance and become the de facto level of abstraction for most of the ASIC designs.
The question is, “Are we at the end of the rope?”  Does there exist another level of abstraction above RTL where designers can design much more complex chips effectively?  <click> To answer this question, we will need to look to the <click> applications, because applications will always drive the direction of technology.  As I have already mentioned, 21st century applications will be ubiquitous and diverging, and the embedded systems will dominate.  Another powerful force we need to consider is the market. <click> The force of the marketplace will drive the profit for these applications and the manufacturing cost they can sustain.
Putting it all together, you have the exponentially increasing transistors on a chip, ever increasing but diverging products, which is really the driving force for all these technology advances.  There is also the fact that every chip start is so expensive that one design has to serve multiple products.  The old story of shorter time to market and shorter product cycle is also not getting better.  The logical conclusion is that there will be plenty of so call “Intellectual property” reuse, where a design company  will buy a piece of design from some IP vendors or borrow a piece of design from an old product.  And not just IP, programmable and reconfigurable IP such as processor and DSP.  After the chip, now considered a platform, is designed and manufactured, there will need to be customization for different standard protocols for different countries and geographical regions, customization for each of the member of a product line, customization for different product generations over a couple of years, and customization for different products.
Getting back to this next level of abstraction, a large portion of the embedded market  will go  up to the next level of abstraction, that of designing with IP and programmable cores.  In fact, they already have, and will continue to do so.