The Inquirer-Home

AMD talks up its first Fusion chip

Still a way off though
Mon Feb 08 2010, 20:00

CHIP DESIGNER AMD has been talking up some of the x86 design features for its first Fusion processor at the International Solid State Circuits Conference (ISSCC) in San Francisco this week.

The upcoming 'Llano' Accelerated Processing Unit (APU) will see the joining of a 32nm silicon-on-insulator (SoI) Phenom II quad core CPU with a DirectX 11 capable GPU on the same die. This is a more sophisticated approach than Intel's at present, in which it simply adds a GPU chip to the processor package and calls that 'integrated'.

amd-fusion-die

According to AMD senior fellow Samuel Naffziger, the company has focused heavily on cutting power usage in the x86 section of the APU, allowing more juice for the GPU and keeping heat generation and power waste to a minimum.

The SoI approach enables core level power gating-to-ground, thereby allowing the use of NFET transistors rather than the bulkier and more sluggish PFET transistors for power gating while negating the need for a special thick metal layer on the die to handle gate supply redistribution.

Another major design change by AMD has been the overhaul of the clock grid design, taking the standard fully populated grid and stripping it down to the bare essentials.

amd-fusion-clock-gating-grid

Naffziger reckons this has greatly reduced the number of final clock buffers, thereby massively lowering the clock grid metal capacitance and allowing for fine-grained clock gating to reduce clock switching power when idle. He admits that this means that a new grid layout will probably have to become part of the chip design process, but says that the massive power inefficiencies and leakage caused by using a fully populated grid means they are no longer feasible in a Fusion type architecture.

Lastly Chimpzilla has created a digital power management module for keeping tabs on power usage, which it reckons tears strips off the more common analogue temperature and current metering methods used for making frequency scaling decisions, which require larger margins of error due to environmental and die-to-die variations.

The Llano chips are set to start sampling in the first half of 2010 and will only be available to OEMs from early next year, initially being offered on the 'Sabine' platform for notebooks and 'Lynx' for the desktop version. µ

 

Share this:

blog comments powered by Disqus
Advertisement
Subscribe to INQ newsletters

Sign up for INQbot – a weekly roundup of the best from the INQ

Advertisement
INQ Poll

Coding challenges

Who’s responsible for software errors?