Why does IBM have a semiconductor group when the biggest advancements in computing power are coming from architectural innovation these days and the lighting to the end of the Moore’s Law Tunnel keeps getting dimmer? The reason is pretty simple: having a better architecture means little if you have no way to realize it. You can have the world’s greatest architectural plans for a new bridge, but it is meaningless unless the bridge is built. What happens if it doesn’t get build? Well then, you’ll just have to use somebody else’s bridge. That’s why IBM does it. The alternatives are not very good.
First, they could fold back into the open PC architecture they created. But then IBM would just be another me-too PC supplier. Plus, it’s a cost and distribution game that IBM doesn’t execute well to. IBM does execute well to a support and technology game like few others. In fact in this arena, it’s the HP’s and Dell’s who are playing catch up. So why follow the enemy who is following you? Then you’ll just be a dog who is chasing its tail.
Second, they could fold into the open Fabless/IP/Foundry. But this is also a cost game targeted mostly at the communications sector who wants lots of technology on the cheap. PICOS buying methods drive them to force vendors to equality, eliminating all differentiation. They want 30 flavors of vanilla so they can play everyone off against the other. Fabless companies have fantastic targeted marketing. They are demons at taking a mind boggling array of IP pieces and finishing off the customer’s puzzle just as they’re ready to ship. Now, you’re chasing your customer’s tail and I don’t need to tell what inevitably happens when you’re following that end of the customer. You won’t find IBM here. They want to be serving up the meal and leaving the mess for somebody else to clean up.
So, IBM’s strategy is to have a semiconductor operation that realizes their architectural vision. What are the tactics? Strategy is usually very easy: easy to do and easy to get wrong. Without tactical acuity, no strategy will ever work. It’s like looking at the moon and making getting there your strategy. Without a well thought out tactical plan, it’s no more than a dream.
The weakest link in the model is DFM. It used to be that meeting timing was the biggest failure mode for new designs making it to production. While it’s always an issue and it’s still a big issue, failure to yield has been the biggest failure mode in the nanochip era. If you fail to meet timing, you can’t sell the chip. If you fail to meet yield, you can’t AFFORD to sell the chip. So all chips eventually meet timing. Plus the EDA companies have done a fantastic job of eliminating this scourge of the nineties.
Failure to yield is much more insidious: you can ship product on the expectation that you will fix the yield issues. However, you will be forging the chains that tie you to losses. The alternative is not to ship, but then you miss the customer’s design-in window. Most people choose the former and pray they can catch up. The consequences for the fab manager in this situation is that management will probably outsource to a foundry next time around and you’ll be out of a job. So, there is lots at stake in this game.
The key to high performance computing from a chip making perspective is predictable success, fast and low cost design turns. But remember that in design low cost is equal to fast turns and fast turns are equal to predictable success. Predictable success is a DFM issue. If you are good here, you should have success all the way from design to customer application, as this is the weak link in the chain today.
So tactically, IBM approaches this problem by focusing on being the best in four areas: Predictive models, Optics optimization, RETs*, and Design Rules. Little else matters. For example, IBM’s Tim Farrell refers to modern lithography technology as “Programmable Pixilated Illuminators.” His title: Distinguished Engineer, Computational Tech. His world is about getting what the design architecture says should happen in the system to be possible because the patterns have been realized in silicon. So you can see that IBM’s strategy and tactics for DFM are critical to aligning manufacturing with sales and marketing . . . and ultimately to customer satisfaction.
Barrier to entry is another advantage that IBM gains from its semiconductor strategy. Because they control who gets their hands on the chips through the entire process, competitors have a huge moat to cross before they can gain access to their technology. In fact, because so much of IBM’s end business is based on leases, they hold ownership of a product from womb to tomb. It’s akin to the Coke formula that’s been keep secret for more than a hundred years.
So, alignment to market and barriers to entry are the primary motives for IBM’s DFM and semiconductor strategy.
* RET: Reticle Enhancement Technology – From 2007
By G Dan Hutcheson
Copyright © VLSI Research Inc. All rights reserved.