Ex-AMD VP questions logic of chasing process shrink

One of the most dedicated chases in the IT industry is the one that Intel, nVidia and AMD engage in when it comes to process shrinkage. Indeed, with its vast array of fabrication plants, Intel's entire mantra is based on the idea of Tick-Tock: Shrinking a known technology, before inventing a better way of doing things. But does this strategy make sense for companies like AMD on such a broad range of products? KitGuru dons a slightly baffled look and heads into the world of lithography.

Every day, we all cross roads. When we do so, it's with the knowledge that our chances of getting to the other side, safely, are close to 100%. If you knew that a specific route to your destination was much quicker, but it meant that you need to cross a road where the chance of survival was close to zero – would you take that route?

The non-suicidal among us would avoid the ‘path of near certain death'. But the same logic does not hold true in processor manufacturing market.

SuVolta is backed by huge venture capital firms that have bought into the company's ethos that reducing the power requirement for a circuit is almost as important as process migration. Indeed, it aims to reduce the power required by as much as 50%, while cutting leakage by 80% or even more. The cornerstone of this aim is the Deeply Depleted Transistor technology being developed by SuVolta.

OK, so the concept is interesting enough, but what kind of people are running SuVolta?  Well the board of directors includes ex-ATi CEO Dave Orton and the technical team is riddled with experts from both Intel and AMD. So these guys have half an idea of what they are doing.

Going back to the failure rates associated with bring a new process/technology to market, there are huge challenges to a company like AMD. Research by companies like VLSI shows how tough it can be to work with a new technology. It started by looking at the 1-2 year period before a company decides to ramp up to full/live production – and the way this impacts costs:-

Massively increased cost per chip in the first 12-24 months of working with a new process (which can be up t0 3x what it will be in years 3 and 4 – once the technology matures)

At the same time, what you're producing is likely NOT to work, with up to 100% of new chips failing on the initial pre-production run, when the new process is first tested

In fact, the very first 12 months in the life of a new process is where yields rarely go above 10% – with ‘working chips' only passing 50% during the 3rd year of a process being used

Now it's clear that whenever a company like SuVolta begins pushing its technical specialists forward to present new research, that they have an agenda. That agenda includes selling its own DDC technology. Fine. No problem. We understand that. But are there any general lessons to be gleaned from this analysis?

Here's a question that, if we were share holders for a company like AMD, that we would be asking right now: For how long does it make sense to try and compete, across a huge range of markets, when each launch needs the latest/greatest/most expensive new process technology ever seen? the actual numbers will vary from company to company, but there are organisations out there who can deliver a new product for $10-30 million – whereas some of AMD's projects could touch the $100 million mark. Ex-AMD VP Nick Kepler is one of those who believes AMD needs to look at a different option.

Many experts believe that what we have experienced, so far, with the introduction of new processes etc has been ‘smooth sailing' compared to what will happen when we hit 20nm or smaller. It's not that the 3rd party FABs cannot deliver – they are buoyant about their abilities in this area – but how mush cost can a company absorb and still come out fighting?

KitGuru says: You only have to look at the way AMD competes in the graphics market – and compare that to the way it competes in the high-end CPU market – to see the difference. One is a fairly evenly balanced combat zone and the other is extremely toxic. Using what AMD would probably consider quite antiquated technology, Apple has managed to build mega-business in the tablet space over the last two years or so. How many new $100m projects does AMD need to be kicking off in order to becomes a Fortune 500 company? Now there's the question.

Comment below or in the KitGuru forums.

Become a Patron!