Px32s

Circuitsoft

Any chance of releasing the prop as a synthesizable licensed core a-la ARM?

jmg

The Prop 1 does not exist in that form, but the opcodes are published, so you could do a Soft-cpu of at least some of the Prop. Some of the analog pin stuff, would struggle in translation ....

A logic chip that would swallow this, would cost more than a Prop (& likely more than Prop II) - so it is a niche answer....

But it might allow more pins, or better timers, or easier off-chip expansion, or more (or less) COGS to suit... ?

Some split between Hard-COGS and Thread-COGS could make sense in a HDL version.
The original die simply cut-pasted 8 copies, but the Logic costs of that, might drive a more frugal mix like 2-4 hard COGS, plus 6-4 (or ++) time-sliced ones (ie share the opcode engine, and slice RAM access )

Circuitsoft

What I'm imagining is a 4-cog prop sitting as a peripheral in the address space of an ARM MCU (Freescale i.MX 5x?), with host-cpu access interleaved between each cog access.

pedward

I had this same discussion with Chip last night or the night before. He said the Prop 1 is written in AHDL and it would be simple to convert to Verilog.

That being said, I thought it could be a neat way to make some more money, but the logistics of making it happen are not trivial. You would have to write up comparative analyses of the Prop 1 vs other microarchitectures and decide on a price, plus you have to support the chip at a macro level that he doesn't have to do now.

It isn't a market or decision that you get to lightly, there is a lot of competition in that area and it's tough to realize value from IP in that sense. People are willing to pay $5 for a $1 chip, but getting them to pay $2 per licensed device, that's a tough sell.

It's the same mentality as gas prices; people are attracted to gas that is 3 cents a gallon less at a dive gas station instead of 3 cents more at a nice station with no waiting. When you are buying 15 gallons that is 60 cents, hardly a lot to worry about, but people go for it.

Circuitsoft

NetSilicon used to make ARM9 processors with a pair of these PICs, running at 400Mhz, as peripheral controllers. Nowadays, I think they're embedding a rabbit6000 (high-speed Z80) instead.

[EDIT]I take that back; they're putting the PICs in R6k processors, and getting out of the ARM business[/EDIT]

jmg

Circuitsoft wrote: »

What I'm imagining is a 4-cog prop sitting as a peripheral in the address space of an ARM MCU (Freescale i.MX 5x?), with host-cpu access interleaved between each cog access.

Sounds useful - the MachXO2 series could perhaps manage that, with some limit on total RAM.
They have a strangely flat price curve (Small parts cost much more than their % scaled logic), but the very largest part is not a bad price at $11.55/100+ (next size down, is 63% of the Logic and 28% of the RAM, for 80% of the price)

Others already offer Asymmetric Dual-cores, and a M4+M0 LPC4330FBD144, with 264K RAM, is $8.97/100+

There is even a vapor-ware? part I found here, claims to be an ARM+8051 split ?
http://www.icbase.com/psen/products/products.asp?dept_id=1123

Circuitsoft

OMAP5432 is a Dual-core Cortex-A15 with two Cortex-M4s, a C64+ DSP, and a PowerVR 3D accelerator.

While the M4 is somewhat DSP capable, I still think the Prop would probably make a better I/O or power controller, which is what I think they're there for.