TOMI™ Briefing Agenda
October 20, 2007
Russell Fish III
Computer industry is at a dead end.
There is no question that the computer industry must make a dramatic change.
The only question is whether TOMI will be the change.
TOMI architecture appears to be first with the solution to both power and memory bandwidth problems.
Why is the TOMI™ architecture special?
- The first CPU designed specifically for embedding on main memory.
- Small enough to fit on a DRAM. 160,000 transistors vs. 270million transistors for Pentium
Implications of very small CPU
- able to physically fit on DRAM
- even able to physically fit multiple cores on DRAM
- able to be routed with 2 layers of metal
Implications of CPU on DRAM
- "Power wall" is crushed
- - Driving teeny wires vs. driving huge pins and capacitors
- "Memory wall" is crushed
- - Practically no limit to size of on-chip bus. 8192 vs. 128 bits
- CPU architecture substantially simplified by CPU and memory together
- - Cache not needed. 2/3 of transistors and 3/4th of power is used by Pentium cache
What are the market implications?
Memory and Processor Markets Merge
1. Intel $45B
They are dead in two dimensions (James Reinders)
- No Pentium faster than 4Ghz
- Increasing speed 15% increases power 75%
- Multicore is only option
- CPU speed already outrunning memory bus speed
- Multicore just makes it worse
INTEL invented DRAM, but no longer makes DRAMs.
INTEL strategy is to drive logic fabs to more sophistication at huge expense, freezing out competition.
(This strategy becomes no longer effective.)
2. AMD $13B (Intel-lite)
3. TI $13B (DSP)
Nine out of 10 DSPs are integer
4. Memory makers who don't have TOMI
DRAM and Flash makers who have TOMI
- Adding less than 1% to any DRAM, SRAM, or Flash memory offers customers 1 or more 32-bit CPUs.
- Memories continue to work as memories.
Memory business is high volume/no margin.
Processor business is low volume/high margin.
Intel $45B revenue $8B profit annual run rate.
TOMI™ Market Breadth
Overview of the breadth of market coverage from the smallest to largest.
From smallest disposable computers to largest data mining supercomputers
We Do Not Do
Floating Point Supercomputers (2% of commercial market)
Weather forecasting, Global warming, Weapons modeling