News Focus
News Focus
icon url

wbmw

02/04/11 9:47 PM

#98696 RE: chipguy #98658

ARM could have done quad core at 350 nm with ARM7 cores if
they had wanted to.


Looks like the hard macro size for a dual core performance optimized A9 core on 40nm is 6.7mm2, which allows 2GHz operation @ 1.9W according to ARM. So it would be around 13.5mm2 for quad core.

http://www.arm.com/products/processors/cortex-a/cortex-a9.php

Just doing the crude math, that would come out to be (not including I/O or other SOC units):

13.5 * (350/40)^2 = 1034 mm2

How big was the reticule on 6 inch wafers, again? :-)

By the way, if you really wanted to start with something modern, but small, there's always the ARM A5 core, which is 0.53mm2 including L1 cache:

http://www.arm.com/products/processors/cortex-a/cortex-a5.php

Of course, it only offers 1600 DMIPs @ 1GHz, compared to almost 5k per core for the A9, but die size does come with tradeoffs. If you could gang together 4 cores, it would be somewhat larger than 2.12mm2. Call it 2.3mm2.

2.3 * (350/40)^2 = 176 mm2 (still not counting I/O or other units)

A quad core A5 on 350nm would be a bit more modest, but I bet it would clock under 50MHz, have the performance of a 386, and dissipate at least as much power.