A Proposal for Resolving the P2 Power Issue
Dave Hein
Posts: 6,347
If you look at the P2 Spec at http://www.parallaxsemiconductor.com/Products/propeller2specs you will see that the P2 had the following features:
If the earlier design for the P2 is doable then some of the other features could be considered for addition if they still allow the chip to meet the power goals.
This approach allows Parallax to move forward with the P2 design rather than regressing back to the P1 and adding features to it.
P1 Core 4-port cog memory 20-bit multiplier 126K RAM 2K ROM 32-bit Multiply/Divide Cordic Engine PTRA/PTRB INDA/INDB 256-long CLUT/FIFO PTRX/PTRY Data Cache 265 Instructions 128-bit Hub BusI think this spec was posted about 2 years ago. Since then, Chip realized that if he got rid of the huge bus supporting ADCs for every pin he could add more hub RAM, which is probably a good thing. However, it then opened the flood gate for lots of new features, which consume real estate and require power. The major new features that were added were:
256-bit Hub Bus 4 tasks 128K More Hub RAM hubex 4 Instruction Caches serial I/O Pre-emptive threads 100+ Special InstructionsThere are various proposals to go back to the P1, and add some features to make it a more useful chip. I think this is the wrong approach. I propose to Parallax that they go back to the P2 at the point when the big bus was removed and the number of ADCs was reduced, but before the extra 128K of RAM was added. Evaluate the power requirements of that chip, and determine if it is doable at 180nm.
If the earlier design for the P2 is doable then some of the other features could be considered for addition if they still allow the chip to meet the power goals.
This approach allows Parallax to move forward with the P2 design rather than regressing back to the P1 and adding features to it.
Comments
More RAM
WIDE RAM bus
ICACHE x 4
Pre-emptive task switching
More instruction customization
I might be wrong, but believe that hubex 1.0 was available prior to the DAC bus removal. Single threaded hubex wasn't a huge complexity adder, it was the 4 256bit wide cache lines that added 9000 flip-flops or transistors (can't recall) to each COG.