PragmatIC -- Logic ICs printed on flexible plastic

[Rick C. Hodgin]

This is the most exciting thing I've seen in semiconductors since the
early 1970s when scaling and yields began to reach high enough to pro-
duce the first truly versatile products in an integrated package, the
early 4-bit, 8-bit, and 16-bit CPUs.

These are flexible, printed ICs using a simple manufacturing processing
technology that costs 100x to 1000x than traditional billion+ dollar
silicon-based facilities. They are targeting the sub-$0.01/unit markets.

Their primary customer target is to integrate electronics and the
ability to have logic on products whose primary purpose is not to be
an electronics component, such as the animated cereal boxes seen in
the movie Minority Report.

The recovery on a fab investment is in the two-year timeframe:

At 7:13:
www.youtube.com/watch?v=WchJR6qDq9o&t=7m13s

This is the third generation of the technology, and from gen1 to gen3
they've seen a 10x reduction in circuit area in two years, and are
projecting another 10x in a similar timeframe:

Gen1: [-------------------------------------------]
Gen2: [--------------------]
Gen3: [----] With Gen3 optimization, PlasticARM is ~16K gates.

At 12:10:
https://www.youtube.com/watch?v=WchJR6qDq9o&t=12m10s

They are in now volume production for 2018 customers, have met their
goals for cost per unit, and have exceeded their expected performance
and power estimates.

At 10:07:
https://www.youtube.com/watch?v=WchJR6qDq9o&t=10m7s

-----
In the keynote Q&A, the CTO mentions the technology operates up to the
~100K gate and range economically, and from the 2 KHz range up to about
1 MHz. The product demonstrated in the interview video shows a 32-bit
ARM1 core built around more modern Cortex-M logic, with some custom
optimization added by ARM over each generation, making it a low-gate
count product, but one which demonstrates a true SoC product with
integrated memory, CPU, I/O, and more.

Interview with PragmatIC's CEO, Scott White, and Kris Flautner,
VP Technology at ARM (ARM is now an investor in this technology),
and also PragmatIC's VP of business development, Joao De Oliveira:
https://www.youtube.com/watch?v=01y6bR6ETpA

Keynote at IDTechEx 2017 with specific details:

Scott White, CEO PragmatIC:
https://www.youtube.com/watch?v=WchJR6qDq9o

-----
In summary:

Integrated logic, memory, and I/O circuits on flexible plastic.
Scalable from ~10 to ~100K gates economically today.
Operational speed from 2 KHz range up to around 1 MHz today.

Demonstrated proof-of-concept into the GHz range, but do not
plan to offer that commercially in the near future, as their
target customer is the extreme low-cost-per-unit market.

Future scaling is projecting a 10x reduction in size, making
1M+ gates possible, and higher frequencies, with the tradeoff
being indicated as in silicon: high performance or low cost.

They already have a 32-bit ARM1-style SoC working.

In volume production. Expecting to ship ~1B units in 2018,
with a long-term goal of ~1T units per year in the future.

--
Rick C. Hodgin

[Rick C. Hodgin]

On Saturday, December 16, 2017 at 4:35:58 PM UTC-5, Rick C. Hodgin wrote:
> These are flexible, printed ICs using a simple manufacturing processing
> technology that costs 100x to 1000x than traditional billion+ dollar
> silicon-based facilities. They are targeting the sub-$0.01/unit markets.

Should be 100x to 1000x less than traditional...

> Interview with PragmatIC's CEO, Scott White, and Kris Flautner,
> VP Technology at ARM (ARM is now an investor in this technology),
> and also PragmatIC's VP of business development, Joao De Oliveira:
> https://www.youtube.com/watch?v=01y6bR6ETpA

In this video, the interviewer asks about the complexity of the ARM1
CPU and their long-term goals, such as integrating graphics. ARM's
Kris Flautner replies that his goals are initially to get some kind
of general processing in a flexible package like this (the equivalent
of a Cortex-M0 or even "M minus one," and then to move forward on the
substrate to find targets for the technology through customization.

Their goals are low-cost products and extremely wide integration.

--
Rick C. Hodgin

[Rick C. Hodgin]

On Saturday, December 16, 2017 at 4:35:58 PM UTC-5, Rick C. Hodgin wrote:

> This is the third generation of the technology, and from gen1 to gen3
> they've seen a 10x reduction in circuit area in two years, and are
> projecting another 10x in a similar timeframe:
>
> Gen1: [-------------------------------------------]
> Gen2: [--------------------]
> Gen3: [----] With Gen3 optimization, PlasticARM is ~16K gates.
>
> At 12:10:
> https://www.youtube.com/watch?v=WchJR6qDq9o&t=12m10s

It uses approximately 10 layers using some traditional semiconductor
processing steps, and some new ones. It is part of the name of their
company, "pragmatic" which compromises on what's best for the tech.
There are dielectrics, and the equivalent of poly-si layers through
new materials development:

At 5:58:
https://www.youtube.com/watch?v=01y6bR6ETpA&t=5m58s

Registration between each layer is one of the challenges, as the
plastic might move or expand with temperature variations, etc.

--
Rick C. Hodgin

[Quadibloc]

On Saturday, December 16, 2017 at 2:35:58 PM UTC-7, Rick C. Hodgin wrote:
> This is the most exciting thing I've seen in semiconductors since the
> early 1970s when scaling and yields began to reach high enough to pro-
> duce the first truly versatile products in an integrated package, the
> early 4-bit, 8-bit, and 16-bit CPUs.

> Integrated logic, memory, and I/O circuits on flexible plastic.
> Scalable from ~10 to ~100K gates economically today.
> Operational speed from 2 KHz range up to around 1 MHz today.

It is good that there is the ability to do this, as there are lots of
applications for such a technology.

But I guess I'm parochial; 100,000 gates at 1 MHz is not enough for a powerful
CPU, so it doesn't get me all that excited. However, even a 4-bit CPU is enough
for a pocket calculator - and some of those did things that CPUs normally did
not do, at least in the old days, like log and trig functions, and so with
microcode one can do a lot.

Imagine a textbook on System/360 programming with a System/360 printed on one
page! That would be both exciting and perverse; and if it was a textbook on x86
programming, a slow x86 in the book would serve little purpose, as real x86
systems are ubiquitous.

Computers are so cheap in the *used* market that a less expensive but less
powerful computer, to be used _as_ a computer with a monitor and a full
keyboard, isn't a rational purchase. But less expensive and less powerful
computers have been a roaring success in things like smartphones.

Smartphone companies are working on bendable displays. So far, that technology
is expensive. And the new disruptive technology _du jour_ is smart watches;
thin, flat, but big doesn't fit that.

Even with existing technology, there are a lot of children's books that have a
little plastic computer thing along the top, to play music or animal sounds
along with a story book.

A book could have a flap that folds out, providing a solar-cell powered
scientific calculator, or chess-playing computer, or so on. But that's a
thumpingly unimaginative application for the technology - existing technology
can already do that, as the cover of a book doesn't have to be bendable.

Of course, if this technology plus display technology gets really cheap, we
could have advertising flyers that play movies. Again, though, not an exciting
application, more like an annoying one.

Still, even if I can't figure out what it would be really good for, I'm sure
their customers will, and many exciting new devices will be coming our way.

John Savard

[Bruce Hoult]

On Sunday, December 17, 2017 at 10:16:53 AM UTC+3, Quadibloc wrote:
> On Saturday, December 16, 2017 at 2:35:58 PM UTC-7, Rick C. Hodgin wrote:
> > This is the most exciting thing I've seen in semiconductors since the
> > early 1970s when scaling and yields began to reach high enough to pro-
> > duce the first truly versatile products in an integrated package, the
> > early 4-bit, 8-bit, and 16-bit CPUs.
>
> > Integrated logic, memory, and I/O circuits on flexible plastic.
> > Scalable from ~10 to ~100K gates economically today.
> > Operational speed from 2 KHz range up to around 1 MHz today.
>
> It is good that there is the ability to do this, as there are lots of
> applications for such a technology.
>
> But I guess I'm parochial; 100,000 gates at 1 MHz is not enough for a powerful
> CPU

1 MHz is not fast, but that's enough gates for a decent CPU: 68000, for example, or yes an early ARM before they got all the cruft. ARM7TDMI, for example: 74000 gates, including a fast multiplier. ARM11 added SIMD but needed 111000 gates. I'm not sure how many gates a Cortex M0 has. I guess fewer as it doesn't have the multiplier, and supports only a single instruction encoding.

If you don't want to get sued then a low end RISC-V implementation would be a better bet now. There are some on github using just over 300 6-LUTs (plus I assume prebuilt FPGA ALU and register files and muxes). In gates I'd expect an RV32E to be very similar to ARM2, or maybe a little fewer.

> Computers are so cheap in the *used* market that a less expensive but less
> powerful computer, to be used _as_ a computer with a monitor and a full
> keyboard, isn't a rational purchase. But less expensive and less powerful
> computers have been a roaring success in things like smartphones.

Yes.

I find it very hard to justify anyone buying a Raspberry Pi along with SD card (costing as much as the Pi), and new monitor, keyboard, mouse for use as a general purpose computer when you can get a much faster used Core2 Duo full system for the same money -- if not for free these days.

If you want to embed it into something, or interface it to custom hardware then the Pi is great. But completely pointless as a general purpose PC.

[paul wallich]

On 12/17/17 3:47 AM, Bruce Hoult wrote:
> On Sunday, December 17, 2017 at 10:16:53 AM UTC+3, Quadibloc wrote:
>> On Saturday, December 16, 2017 at 2:35:58 PM UTC-7, Rick C. Hodgin wrote:
>>> This is the most exciting thing I've seen in semiconductors since the
>>> early 1970s when scaling and yields began to reach high enough to pro-
>>> duce the first truly versatile products in an integrated package, the
>>> early 4-bit, 8-bit, and 16-bit CPUs.
>>
>>> Integrated logic, memory, and I/O circuits on flexible plastic.
>>> Scalable from ~10 to ~100K gates economically today.
>>> Operational speed from 2 KHz range up to around 1 MHz today.
>>
>> It is good that there is the ability to do this, as there are lots of
>> applications for such a technology.
>>
>> But I guess I'm parochial; 100,000 gates at 1 MHz is not enough for a powerful
>> CPU
>
> 1 MHz is not fast, but that's enough gates for a decent CPU: 68000, for example, or yes an early ARM before they got all the cruft. ARM7TDMI, for example: 74000 gates, including a fast multiplier. ARM11 added SIMD but needed 111000 gates. I'm not sure how many gates a Cortex M0 has. I guess fewer as it doesn't have the multiplier, and supports only a single instruction encoding.
>
> If you don't want to get sued then a low end RISC-V implementation would be a better bet now. There are some on github using just over 300 6-LUTs (plus I assume prebuilt FPGA ALU and register files and muxes). In gates I'd expect an RV32E to be very similar to ARM2, or maybe a little fewer.

Even those of us who lived through it tend to forget how much useful
work could be accomplished with a 1-Mhz clock. Anything embedded that
doesn't require realtime control. And the packaging becomes so much
simpler and smaller if your circuits can flex. If the clock rates remain
low, I foresee the (re)development of a bunch of low-data-rate
communications protocols. (Once again, for most applications a few
hundred to a few thousand bits per second will do just fine.)

paul

[Rick C. Hodgin]

The 1 MHz clock and 100K gates are cost barriers today based
on reasonable development. They are expecting a 10x reduction in
feature size, resulting in higher clock speeds and more gates.

Some customers don't want the technology for its flexible nature,
but because the package is so flat. This will probably result in
several-hundred MHz products.

--
Rick C. Hodgin

[bitrex]

You can do a lot of useful stuff without having a full processor on
board. Silego Technology (now owned by Dialog Semi) makes these little
"GreenPAK" one-time programmable mixed-signal processing blocks that can
perform lots of useful functions; they usually just have a couple
configurable timers/oscillators on board, a couple analog comparators,
and say 10-15 functional blocks that can be configured as gates, flip
flops, lookup tables, etc. along with an i2c interface. You design the
hardware in a GUI by selecting and connecting up the functional blocks,
sort of like Multisim.

If the "flexible ICs" could at least be made user-programmable that
opens up a lotta possibilities

<http://www.silego.com/products/greenpak.html>

[Bruce Hoult]

On Sunday, December 17, 2017 at 6:27:13 PM UTC+3, paul wallich wrote:
> On 12/17/17 3:47 AM, Bruce Hoult wrote:
> > 1 MHz is not fast, but that's enough gates for a decent CPU: 68000, for example, or yes an early ARM before they got all the cruft. ARM7TDMI, for example: 74000 gates, including a fast multiplier. ARM11 added SIMD but needed 111000 gates. I'm not sure how many gates a Cortex M0 has. I guess fewer as it doesn't have the multiplier, and supports only a single instruction encoding.
> >
> > If you don't want to get sued then a low end RISC-V implementation would be a better bet now. There are some on github using just over 300 6-LUTs (plus I assume prebuilt FPGA ALU and register files and muxes). In gates I'd expect an RV32E to be very similar to ARM2, or maybe a little fewer.
>
> Even those of us who lived through it tend to forget how much useful
> work could be accomplished with a 1-Mhz clock.

Less than ten years ago I was writing a Java native compiler aimed mostly at those ARM7TDMIs I mentioned, and supporting people writing phone apps and games for them. Most of them were running at 1 MHz, with a few at 4 or 8. They generally had 1 or 2 MB of RAM, though some had only 400 KB.

A 1 MHz ARM is pretty similar to an 8 MHz 68000 for speed. Register to register instructions take 4 cycles on the 68K, so that's faster, but the rest take 8, 12, 16+ cycles.

> Anything embedded that doesn't require realtime control.

1 MHz is plenty for many kinds of real time control. It's no problem at all to run a software PID controller at 100 Hz or at most 1000 Hz for many many applications. You can do that on a 1 MHz RISC no problem .. even with software floating point.

This is commonly done with 8 bit AVR chips, running at 8 or 16 MHz. For example most amateur 3D printers, where you're doing real time control on four axes of stepper motors (one for the extruder speed), plus temperature control of the extruder and possibly the bed.

[Quadibloc]

What application for digital logic would benefit obviously from being thin and
flat?

I've come up with one, but it does not really make me feel excited.

How about paper money that can prove it isn't counterfeit thanks to a complicated
cryptographic protocol?

John Savard

[paul wallich]

If not money, other tokens of value, and also all of those holographic
seals that are supposed to protect/authenticate things but don't because
they can be reproduced easily.

But being thin and not-flat may be an even bigger deal in the slightly
longer term: All those cheapish objects whose form factor is currently
constrained by having to contain and protect rigid pcbs and component
packages of appreciable thickness. Those will benefit a lot from
circuitry that can just be crammed in any which way and needs negligible
mechanical support beyond strain relief for electrical connections.

paul