by Anton Shilov
08/30/2012 | 09:04 PM
A new approach of Advanced Micro Devices to microprocessor design could lower manufacturing costs, decrease power consumption and even compensate for AMD's consistent lag with transition to new manufacturing processes compared to arch-rival Intel Corp.
Die size of chips directly affects not only their costs and yields, but also their power consumption. Therefore, it is crucial to shrink die sizes to limit power consumption and/or even improve performance. In a bid to do this, Advanced Micro Devices will adopt "high-density" and "thin" libraries of elements typically used to design graphics processing units (GPUs) to improve its code-named Excavator and post-Bulldozer family central processing units (CPUs) and accelerated processing units (APUs).
For the APU and CPU products made using 32nm SOI HKMG process technology, AMD's design team used a combination of automated place and route and hand-placed semi-custom design, which reduces power and area somewhat. In future products, AMD will be employing “dense” or “thin” libraries employed by its GPU design teams, but for APU/CPU implementations, to deliver more power efficient designs.
During a keynote at Hot Chips conference, Mark Papermaster, chief technology officer of AMD, revealed that adoption of "thin" and "dense" libraries of building blocks used by GPU designers can reduce die size and power consumption by as much as 30%, which is comparable to an improvement resulting from a switch to a more advanced process technology. Optimized designs could also result in higher amount of execution units within a chip, which directly affects performance and functionality of chips.
Usage of ultra-dense and ultra-thin libraries of elements is an industry standard design methodology well adapted to a foundry model. Cost and power consumption reductions by 15% - 30% are no less than impressive. What remains to be seen is whether specific libraries affect scalability of designs in terms of clock-speeds as today frequencies are still one of the most important factors that affect performance of microprocessors.