Take process, apply more polish, rinse, repeat…

by Josh Walrath

AMD and Strained Silicon

            As mentioned before, it has become known that AMD is utilizing strained silicon for its latest 90 nm parts, and will utilize this also for some of its 130 nm parts.  Utilizing strained silicon will add a few more steps to the chip making process, but due to how relatively inexpensive implementing it is, end users will not see a significant price increase for this technology.

            AMD has transitioned to the 90 nm node in a very smart way.  They have taken their time to make sure that the underlying process is solid, and they have not radically redesigned the Athlon 64 core for the 90 nm node.  By not implementing a new and more complex core, AMD has reduced the risk of going to the 90 nm node.  Because the current Athlon 64 core is very well known to the engineers at AMD, adjusting it to be used at 90 nm is relatively simple as compared to what Intel did with the much larger and complex Prescott core and their 90 nm process.  The Prescott was designed to run at high speeds, and as such the transistors had to run very fast.  To get them to go that fast a lot of current had to be used.  The faster the design needed to go, the more current that needed to be applied.  Intel was hurt by leakage due to the speed requirements, as heat and power soon spiraled out of control.  If Intel had originally transitioned the Northwood core to 90 nm, I doubt that they would have seen the problems they have experienced with Prescott.

            By only optimizing the Athlon 64 core for 90 nm use, AMD has allowed their engineers to concentrate on getting good yields at current speeds for their 90 nm process, and has also allowed them to integrate strained silicon into the design.  This appears in fact to be the approach taken by AMD, as their first 90 nm processors are aimed at the mobile market, and the upcoming desktop processors will be aimed squarely at the midrange (2.0 GHz to 2.4 GHz products).  While it is speculated that AMD will only realize a 10% gain in speed from using strained silicon at this time, further enhancements made at the 90 nm node should allow AMD to ramp the Athlon 64 core to 3.0 GHz within a year, and possibly a maximum of 3.2 GHz on the 90 nm process with the current design.  Strained silicon is a win/win situation for AMD, as it is not overly complex to implement, and the power and heat characteristics of transistors based on strained silicon are significantly better than their standard silicon counterparts.

            AMD has also seen fit to integrate strained silicon into their 130 nm designs.  This is a surprising, but very smart move by AMD.  This allows AMD to get another speed increase out of the 130 nm process with very little cost and essentially no major redesigns.  The current 130 nm parts are running at 2.4 GHz, and even with some extreme overclocking are not able to get to 2.6 GHz and beyond.  AMD is planning on one more speed increase on their 130 nm Athlon 64’s, and this looks to be achieved by incorporating strained silicon into the design.  Even though strained silicon will not turn these into 3 GHz monsters, they will allow for good yields on 2.6 GHz aircooled products.  The extra 10% this technology gives to a design will be enough for AMD to continue competing with Intel in ramping clockspeed and performance up.

            AMD’s 90 nm process now appears very aggressive with the inclusion of SOI and strained silicon.  It is no wonder that the first 90 nm parts are aimed at the mobile market, as well as the mid-range desktop market.  AMD probably has very good yields with this process at the 2.0 to 2.4 GHz range, but just like everyone else in this field, it will take a while for AMD to adjust the mixture to achieve good yields at the faster fast speed bins.  This is why AMD is relying on the tried and true 130 nm process with strained thrown in to make up the parts that will run at 2.6 GHz later this fall.  For the time being, AMD will be utilizing the 90 nm process to produce large amounts of processors at a lower price that will address the demand in the 2.0 GHz to 2.4 GHz market.

            In about six months AMD should have improved its 90 nm process enough that it will go past the 130 nm 2.6 GHz parts that were previously the top end, and continue to ramp up the speed on that process.  Again, by using the same basic Athlon 64 core design, they are removing one more variable to producing a fast part on the 90 nm process.  AMD also looks to have working dual core samples by Q1 of 2005, and have shipping product by the end of 1H 2005.  Having a rock solid 90 nm process for such a product is absolutely imperative for AMD if they expect to compete with Intel and their future offerings.

Next: Future Possibilities

Copyright 1999-2004 PenStar Systems, LLC.