Moore's Law repealed! But does it matter?

I'll start with first the meaning of "Moore's Law." Moore's Law was really a prediction on how rapidly the SC industry would be able to integrate transistors onto one piece of silicon. It was something like "we will be able to double the number of transistors we can integrate on one piece of silicon every two to three years." In fact, that prediction has been so successful that other corollaries have become attached.

Such corollaries as performance and power dissipation have become a part of our drive for technology. I'm not sure when performance was added as a driver, but I do remember when power dissipation was added as I was one who added it. I called it "Genes Law" and it had to do with my observation that the power dissipation to do a DSP function was being reduced at a rate of one half every eighteen months " the same rate that performance was increasing and, at that time, transistors per unit area were doubling. And, it seems that performance, as measured by clock speed, fell off of the trend line in the early 1990s and Gene's Law fell off of the trend line for a couple of years around the 90 nm node.

But, back to Moore's Law. I'd like to address it from the perspective of the systems our customers are designing. Honestly, our customers don't care about process nodes. They don't come to us and ask for a 90 nm part or the latest device on 45 nm. Our customers are asking for a solution to their problem and not a new technology node. And at TI, we are more interested in meeting the needs of our customers than in getting caught up in how far Moore's law will take us down the nanometer trail. No, this doesn't mean we don't care about Moore's Law, but is does mean we have put it into perspective.

Let's start with voltage scaling. Along with driving the technology, the voltage goes with it. We are now specifying operating voltages in the one to two volt range. But, unfortunately, the real world hasn't followed us. Car batteries haven't gone to two volts to make the system easier. In fact they are going the other way. Car batteries are on their way to 48 volts and higher. Alternative energy requires hundreds of volts. My pointwhat fits these applications best are older technology nodes, not newer ones.

Then there is the demand for faster time to market and more flexibility by our customers. It's the older technology nodes that provide both, not the newer ones.

I tell people that my customer wants four things from me:

A device with good enough performance
A device with low enough cost
A device with low enough power dissipation
So they can be first to market

In fact, ignore the first three. My customer wants me to help them "be first to market with a roadmap to cheap." And what I mean by a roadmap to cheap is that they want the lowest priced solution they can get that still allows them to be cost competitive in the markets they serve.

Finally, there is the idea that we can provide more technology than our customers can figure out how to design into their systems. I'll cover that in my next blog: "A Buck a Billion." It won't be long before the cost of transistors will be in the range of one US dollar per billion transistors. In my blog, I will argue that there aren't many opportunities that need that many transistors at any cost.

So, back to the question at hand, "Is Moore's Law coming to an end?" It may very well be moving in that direction. But, I don't see it affecting the system-level designs our customers are building for a very long time. I believe that what will hold up our customers is their ability to figure out what to do with all of the technology we provide them now. It will be fun to watch

What are your thoughts on the topic?

BTW, I am on twitter. Follow me at @nGENEr.