Hacker News Reader

Comment by londons_explore | Hacker News Reader

Nearly any uses for flexible electronics would also be satisfied by sufficiently small electronics such that lack of flexibility doesn't matter.

Eg. rather than having every pixel in your flexible screen be flexible, you make each pixel rigid and have the joints between pixels flexible.

In this case, this design is based on SERV, which uses ~2100 gate equivalents, which in a recent tech node would be 40 um^2. That means you could fit a 10x10 grid of these in a single pixel on an iphone screen.

I really can't think of a use case where a region 1/100th of an iphone screen pixel being rigid would be a problem.

View on HN

Some questions you might consider which would help you to think of some use cases;

1) How would wiring to you processor work?

2) How many flexible compute applications are currently using just really small processors?

3) Given that Pragmatic has raised a lot of money, what was it in their use case that the investors thought would make a better product?

4) Besides flexibility, are there other requirements in this product space?

5) Given that you've just imagined a product with a flexible screen but solid pixels, does this exist on the market? Are there flexible screens on the market? How do those screens choose to implement flex versus the idea you have proposed? What factors might make their choices better (or worse) than the idea you proposed?

I'm not being critical here, I think you start with an excellent starter question which is "Would the requirements be satisfied by sufficiently small electronics such that [the] lack of flexibility [in the electronics] doesn't matter?"

The trick then is to see if you can see how other people who invested time and money in answering either that, or a closely adjacent, question answered it. When you do that you'll get to see what they thought the overall requirements were vs the technology they picked, and perhaps it might inform if the Pragmatic solution would be a better fit or the 'tiny electronics' solution would be a better fit.

I'll be the first to admit that I'm 'weird' in that I really do enjoy going down these sort of engineering optimization rabbit holes to develop a better understanding of what problems various proposed solutions are trying to solve.

> 1) How would wiring to you processor work?

My (relatively limited) experience is that this is what really makes wearable projects obnoxious.

Even if you have a chip with a tiny footprint, you either put it on a breakout board that isn't tiny or you spend twenty hours soldering nearly microscopic bits of magnet wire to it. It's the same for the piles of passive components and peripherals that every project requires, the voltage regulators and smoothing capacitors and power transistors and stuff: You either attach everything to a big PCB or you're faced with a spaghetti nightmare of point-to-point wiring that makes "normal" dead-bug circuitry, the kind you might find embedded in a block of resin for aesthetic points, look like a walk in the park.

Flexible processors don't necessarily solve that problem, but they definitely demonstrate that flexible circuits in general are advancing in useful ways, better signal quality and longer runs and better process yield. The bigger these things get the better they are for replacing that mess of integration spaghetti that I always see DIY wearable projects suffering from.

(I think that industrial wearables typically solve this by concentrating everything complicated down to a rigid brain-box, c.f. smart-watches or those heated jackets that have a socket for a power tool battery in the pocket.)

> or you spend twenty hours soldering nearly microscopic bits of magnet wire to it.

A real chip would only need two or three contacts to vastly outperform thing demonstrated here. These would probably not be soldered, they would ideally be bonded directly to the chip.

Imagine a near-microscopic 4-ball BGA on a bit of flexible PCB, except the PCB material can flex in multiple axes simultaneously.

Isn't that kind of like how RFID labels work? That's probably the most common use case for flexible anything at the moment, and they work great

You can look at die attach[1] which basically solders to pads on the circuit board (no magnet wires required :-)).

[1] "Die Attach Comes to PCBs" --- https://www.eeweb.com/die-attach-comes-to-pcbs/

That article describes traditional chip on board construction that has been in use for decades. Buried at the end it mentions wire bonding as the final step. i.e. pads are up. Flip chip mounting requires careful attention to thermal expansion coefficients. You're not going to do that on any run of the mill cheap board or flex materials.

> The trick then is to see if you can see how other people who invested time and money in answering either that, or

OT, but I've been wondering how one might teach this young. Perhaps LLM-generated business case studies? Other thoughts? Part of the context is generative storytelling might permit intensively overlaying implicit curriculum on to existing learning objectives (eg, it's a chemistry problem, but chosen to also scaffold biology and illustrate supply chains).

> 1) How would wiring to you processor work?

Sure, there's a modulus gap to be interfaced, but flexible circuits have been worked on forever.

The whole point is of this tech is that the transistors themselves are flexible.

But since the transistors they end up with are orders of magnitude worse than what the microprocessor age started with, to me this just shows that this tech is not anywhere close to practical application.

Yes, the entire chip (wiring and transistors) is flexible. Which is something I find kind of amazing. Back in 2008 there was a company in the UK called 'Plastic Logic'[1] that was going to make an e-reader that could be rolled up. Back when organic LEDs were just starting to be possible and this stuff was living on a glass substrate, doing "all" of the circuits in long change hydrocarbons was a pretty revolutionary idea.

My original point was that dismissing the technology out of hand because you imagine you could solve the same problem with tiny ICs is probably premature. Dismissing any technology coming to market because you think it doesn't solve any problem is usually a bad idea because it takes non-zero effort and resources to bring anything to market. As a result, if you imagine what something is irrelevant because there are other proven solutions, then take that as a signal to say "Hmmm, what am I missing here?"

> But since the transistors they end up with are orders of > magnitude worse than what the microprocessor age started > with, to me this just shows that this tech is not anywhere > close to practical application.

This doesn't really track though does it? The "first" microprocessor, the 4004 ran at 750kHz max. Most of the challenge here appears to be heat dissipation as plastic melts at a much lower temperature than silicon, but the chemistry is still interesting.

I completely agree that this isn't going to displace servers in the data center any time soon, but I can imagine applications for an all (or nearly all) plastic computer on a flexible plastic substrate.

[1] Their IP (not the reader though) lives on at https://www.e-pi.com/

> The "first" microprocessor, the 4004 ran at 750kHz max.

The 4004 wasn't useful to power a general-purpose computer as we think of one today, it was made for a 4 function calculator and it's hard to find many examples of it being used in other systems online. It took another 10 years of Moore's Law for the ingredients to come together and microprocessor-powered desktop computers to achieve critical mass.

Look at Table 2 in the (awful, IMO) Nature article. This thing is 10x slower than even a 4004.

> Dismissing any technology coming to market because you think it doesn't solve any problem is usually a bad idea because it takes non-zero effort and resources to bring anything to market. As a result, if you imagine what something is irrelevant because there are other proven solutions, then take that as a signal to say "Hmmm, what am I missing here?"

100%!

But is there reason to think Moore's Law is happening here?

Or did these researchers just print some minimum viable transistors on kapton?

Slower, clock wise. My guess, considering the IPC, is that it could perform better.

The IPC is far worse too - this is a Serial RISC-V machine - Ie. it takes ~~60 clock cycles for 1 instruction.

Ie. IPC will be 30x worse than a microcontroller core, and 200x worse than a desktop PC core.

(couldn't find exact numbers)

Thank you

40 um^2 might cover the logic (although I think your logic transistor count is a factor of three low; and something like this is most likely to be made on a 45 nm or bigger process), but doesn’t cover IO pads. If you’re willing to wire bond directly to a flex circuit you may be able to use pads on to order of 50 um x 50 um (each! Likely need 6 or 8 pads to be useful), but that’s a hell of a process, and you’d have to encapsulate afterwards, adding bulk. If you want to flip-chip mount it’s pretty hard to go under 1 mm x 1 mm for a useful microcontroller, although there’s some stuff out there at 600 um x 600 um or so from memory — but pad sizes under 300 um then bump up your resolution requirements for the flex circuit you’re bonding to.

They are spot-on with the transistor count for the CPU (though I believe it doesn't include all your RAM registers). SERV is a serial CPU that can range from doing calculations just 1 bit at a time up to doing them 4 bits at a time.

https://github.com/olofk/serv

It's pretty incredible considering the 6502 was already considered tiny when it was released with just ~4500 gates vs the 6500 in the 8085 and the 8500 in the z80.

6502 has 3510 Transistors. 4500 gates seam impossible.

Ignoring flexibility and cost/performance, this may be a sign that rapid chip fab turnaround times are possible. These were made by Pragmatic Semiconductor [1], who claim they can make chips within 48 hours and deliver within 4 weeks (likely due to their use of unconventional materials). Traditional silicon fabs, including trailing-edge foundries and TSMC, take 2-9 months. I do wish they'd emphasized this instead of flexibility.

[1] https://www.pragmaticsemi.com/

Yeah but traditional silicon fabs aren't making 12k gate chips.

What's the turn around time on 12k gate chips from a traditional fab?

Just googling really quick: the Lattice Semiconductor LFE5U-12 is a 12k-LUT FPGA (and a LUT is way more flexible than a gate).

So realistically, if you need fully custom digital logic, you'd buy LFE5U-12 instead and program that.

So that's $16 FPGA from widely available distributors (like Digikey) who likely can afford 1 or 2 day shipping.

-------------

Custom chip design for a flex-circuit is interesting, but only if you have substantial analog parts that cannot be easily implemented by an FPGA.

The power profile of an FPGA is very different (worse) than dedicated silicon. Both the background current and cost of a gate switch. There are a lot of situations where that isn't acceptable.

I suspect the power profile of an FPGA is better than this flex-silicon...

True, that's definitely possible. I was giving the habitual answer (that I occasionally need to explain to a client at work) why you need to spend $2m having an ASIC made when "it's just digital logic and we got the prototype going in a month"

They usually care about at least one of - size, power, cost.

At a certain threshold, miniaturization of electronics can become counterproductive for space applications. The amount of radiation received per unit of area remains constant but our transistor density keeps increasing, which means that every individual event threatens to wreak an increasing amount of havoc (e.g. more bits flipped in RAM per cosmic ray). Considering the increasing amount of error correction and redundancy needed to counter this, we may reach a practical floor on transistor density for such domains.

It's also, if not more so, a factor of transistor voltage. 1.1V transistors are less prone to upset events than 0.7V. It's possible (assumption, here) that some 3+ volt circuits are still used for critical components of the system

It would be possible to use much higher supply voltages if silicon were replaced with a semiconductor material having a higher band gap.

The main obstacle that has prevented this until now is that in all high-bandgap semiconductors it is easy to make only transistors of a single polarity, not transistors with both polarities, as required for CMOS logic. For high circuit densities it would be difficult to replace the CMOS logic, because all alternatives have higher idle power consumption.

... which is a huge problem for solar panels ... but why would it matter for microprocessors?

You don't want transistors operating under the influence of outside forces. More than just having a bit flip in your data, what if one of the control lines in the CPU flips and the entire instruction stream gets corrupted... while you're trying to perform orbital maneuvers.

You use multiple processors and a consensus algorithm.

You can do that.

Leading-edge fab nodes are way too costly for this kind of use. Specialty, low-volume chips are the domain of trailing-edge tech nodes, sometimes even at the μm level. Besides as some sibling comments mentioned, contact pads for off-chip wires would get so big as to ultimately take up most of the area, so there would be no real advantage to using the finer nodes.

Most microcontrollers today are using 40-90 nm processes. That's not the micron level at all. Chips that need current-handling capabilities or have weird needs will use bigger process nodes. This is a big part of why automotive electronics use old nodes.

You're thinking of power mosfets, but automobiles have entertainment systems too.

The power electronics are more complicated than MOSFETs, and the entertainment systems use chips on modern process nodes. The main use I can see for MCUs/MPUs on old process nodes is replacement parts for 10+ year old equipment.

The article is about a CPU, not a display technology - there's a big difference in functionality and fabrication between those two things, so your iPhone screen pixel example doesn't really fit this use case.

I think this flexible CPU tech is interesting. If it's possible to build an ADC onto it and monitor flexible sensors, that would open up one kind of possibility, and probably an advantage over chip-on-flex solutions. I'm sure there are many more interesting uses for this.

It's impressive that a CPU can be implemented with this tech, but interesting things can be done with far fewer gates.

You would need to connect wires to that little 40 um^2 mote to do anything, though, which in practice makes the rigid places needing strain relief a lot larger.

It also makes the amount of strain proportionately smaller.

We're talking about a 3mm bend radius here, so there's a few orders of magnitude to work with.

Not as many as you think, because you need to have targets bigger than the ~20um wirebond wire to solder to... and the little wire and bond pad are not very tolerant of strain at all.

Bonding wires to the thing in the first place easily increases its size by a couple of orders of magnitude.

What are the fab requirements of the two techs? If the ffc version can be manufactured with as basic tech as ffc, then that is huge.

Also bonding small rigid things to flexible things is never actually the same as a flexible thing, in several different ways.

These are not equivalent even if you can manage to use either one for some use cases by accepting various compromises.

> you make each pixel rigid and have the joints between pixels flexible

Alternative 1: Assemble about 3 million individual rigid pixels of an iPhone screen on a flexible substrate, keeping the gaps flexible.

Alternative 2: Produce a single flexible screen piece, requiring no per-pixel assembly.

Which alternative, to your opinion, is likely to cost less?

>Nearly any uses for flexible electronics would also be satisfied by sufficiently small electronics such that lack of flexibility doesn't matter.

we make microelectronics as complex as we can because 128 bits would be better than 64 bits, and floating point, and parallelism, pipelining, speculative execution, and caches, etc.

we already make microelectronics as small as we know how, because speed of light makes a difference, and power dissipation also.

and our chips, already completely consumed by the tasks we give them, are too big to fit your definition of flexibility not mattering.

So-ooo-o, if our chips were flexible it would solve the problem we are both trying to solve, which will not be solved your way, except in some uncertain future.

So, nothing particularly interesting here? When I first saw 'flexible' I immediately thought about balancing a chip's specifications, not its material flexibility!

Flexible circuits are interesting and worthy of discussion.

Really, the whole process here is fascinating to me. There's been a lot of progress in flex circuits over this recent decade.

None of it is electrically or computationally new. It's 1980s tech from a computation perspective. But mechanically??

Being able to weave circuits seamlessly into clothes, tapestry, and such is pretty cool. If only for the cosplay / costume designers but that's still a pretty / beautifully kind of display (especially with a few fiber optics to move lights around).

One of the interesting electro-mechanical issues is that flex circuits are necessarily thin, making grounding / return currents exceptionally consistent. On the downside however, solid planes / ground fills are bad for flexibility, so you apparently need to make a ground-grid instead of ground-fill.

Very interesting tech overall. Even if it's applications are quite small right now.

It's funny, I thought the exact opposite.

> Flexible... isn't that the point of any central processing unit, to be able to handle many differing types of work?

Oh, pliable? that's cool, I wonder how that works?

I agree, this CPU has double the transistors of a 6502, and runs at 1/20th of a 6502 speed. A silicon product will be faster, the size of a printed dot and use less power.

No word either on how many bending cycles the product supports.

Still, it's early days, and flexibility might have some use.

Don't you have to put those BGAs on some kind of pad?

Reply on news.ycombinator.com