Integrated Circuits on Kickstarter?

[devinatkin]

Hello I'm a silicon designer at the University of Calgary. I've been floating the idea of running a Kickstarter for some of our works. Well people generally seem to think of chip development as prohibitively expensive except for extremely large mass manufacture that's not true if one is willing to operate in older process nodes. Quantities are as low as ~35,000 to get the chip prices down to ~$10/chip depending on the details of the process, at engineering quantities of 40 chips it's only $100-150/chip for some processes.

I'm thinking of floating the idea with my supervisor for us to do some of this in order to promote our lab and wanted to get an idea of people's feelings towards this.
I'm leaning towards doing something like a 100% open-source design image sensor design intended for use with Arduino or very small processors. With a simple friendly readout and some moderately good specs. It would be to help get more of our research out into the real world where people could take advantage of it.

I operate in a bit of a bubble where I love the idea so I wanted to get people's reaction to the idea of supporting custom silicon sensors on Kickstarter.

Comments and critique please, tell me if it's a good idea or if my bubble needs bursting.

[ataradov]

There are cheap image sensors out there. Being open source does not really add much. There is still no way anyone could realistically use that source. Also, I don't even think there is a free way to access final manufacturing databases. What are you going to open source? Just the digital part?

It could be a cool experiment, but I don't see it working at all.

[devinatkin]

That's a fair point. I know there are a few cheap image sensors, but none of them I've seen have good documentation. My thoughts would be to make the full schematics with transistor sizing open source. I'd have to check with the manufacturers to make the layout files fully open source or complete the project in software like Magic which isn't as process dependant and is free to use, can't stop me.

Personally, I want to make the sensor because I want to create a unified measure of performance from which I can compare people's works showing how their performance compared to a standard baseline sensor. I'm a bit of a documentation Nazi and I'm not a fan of how a lot of papers don't give a good idea of the tradeoffs with their techniques only focussing on their advantages.

From a hobbyist standpoint, I see the value is having a sensor in a more approachable form factor with a simple instruction set that can be operated through a platform like an Arduino is the advantage.
Our lab does do WDR and some high-speed research so I'd considered looking into seeing about giving it some specs that were a bit more impressive than what one typically gets.

Thank you for your feedback from outside my little bubble.

[ataradov]

You probably need to be more specific. What kind of image sensor are we taking about? What parameters do you expect to achieve?

Also, those prices seem reasonable for older processes, but typically that does not include packaging and bonding. And for samples that can easily double or triple the price.

I guess what I'm trying to say is that we need more details to give reasonable feedback.

[devinatkin]

Fair the $10/chip number I came up with was based on both the silicon and packaging costs we got for a project we were looking at commercializing more seriously. For the engineering run of the initial chips we'd bond them ourselves in the lab to confirm functionality before doing the full wafer runs.

It would be a monochromatic image sensor (Colour filters would add way too much to the cost) likely with high-speed readout, limited by the system it was hooked up to. I'd love to be more specific, but my hunt for general feedback is so that I can justify spending the time it would take to develop the sensor fully and do the full simulations to confirm the functionality. If you say well if it did VGA at 50,000fps I'd say okay that's a target where this might be a practical thing to do.

This is mainly an idea that's been floating around in my head as a cool thing to do at this point. The goal would be to keep the price as low as possible keeping the cost per unit as low as possible unless the specs people were interested in required it to be substantially better.

[dcarr]

The high speed aspect would be very interesting to me.  Might also couple well with a larger process node where you want to intercept more photons?

[RoGeorge]

Not sure what you mean by "Arduino", but to handle a VGA stream of 50k FPS you'll need to transfer a few gigabytes per second.  You'll need an FPGA to read such a fast VGA camera.  Also, I doubt a 50k fps chip is a one man job.  Even as a team, you'll need many iterations to get that right.

The only one-man-show chip I know, other than the sea of academic/research prototypes, is the Propeller microcontroller.  Not sure if kickstarter was invented during Propeller 1, and not sure if it ever was a kickstarter for the current Propeller 2, but AFAIK both of them were made in small batches, mostly by a single person.  Very interesting concept and architecture for a microcontroller, never seen before something like Propeller.

On short, it's possible to get custom chips manufactured (in older technology) for only a few thousands of $, but unless there is a new and closed design, you'll find cheaper on the shelf similar products.

OTOH, if you'll have a 50 kfps VGA chip for $10, it'll sell like crazy! 

[ataradov]

Yeah, 50K FPS is not a realistic thing. KickStarter like that would automatically go to the dodgy technology section. You would need to have a proven track record to convince me to read past the title.

At 50K FPS and $10, I don't care how convoluted and poorly documented your interface is.

[devinatkin]

I shoulda been clear on the 50kfps comment. That was a comment along the lines of pivoting from the simple design concept. It's absolutely not practical for an Arduino controllable design, a decent FPGA would become a necessity to interface with the chip. Datarates being one thing that clever design can't escape lol. I should be a bit more careful about replying while talking on the phone. 50K at VGA would be doable to implement but that design would not be open source. Also while I'm relatively confident I could get those specs in an old process node the other specs would suffer heavily (really noisy picture), and ataradov would absolutely be right to really skeptical of the sensor cause yeah using traditional image sensor design methods it would be complete horseshit.

I'm treating this as a bit of a customer validation exercise for me so thank you to everyone who comments.

[OwO]

Yes it's very difficult to match the image quality of commercial sensors, so you must differentiate your product some other way, like frame rate.

Do you have the capital for the first few prototyping runs? If not, get that sorted first. No, don't even think about VC funding or other kinds of fundraising. I mean actually go do something else for work and rack up some savings first. Otherwise it's easy to end up like most kickstarters, which is you've spent all the money and still have nothing to show for it, leaving angry backers and threats of lawsuits. If you pay for the development out of your own pocket then at least you can still refund backers if it all goes wrong.

If you think that's asking too much, think of it this way. If you don't believe in your own product enough to invest your money in it yourself, why expect backers to collectively put in $200k? Always be your biggest investor yourself.

[RoGeorge]

I'm leaning towards doing something like a 100% open-source design

Open source means it can be copied (legally or illegally) by somebody else who would fabricate your own design faster, or better, or much cheaper than you.

I'm all for open source, but open source designs are not for commercial products.  For example, the Propeller 1 microcontroller is now open source, but that is only because the author now have Propeller 2 for sale.  In the beginning, the Propeller 1 was closed source, and it was made open source only at the end of its commercial life.

Any commercially promising product/idea will be cloned instantly if it's open source, or, if it's closed source, it will be re-engineered soon.  Patents and copyrights won't help much in protecting a design, unless you have the money or the power to enforce that.

[wizard69]

I'd suggest that if you wanted to do this I'd look into demand in the sectors that you intend to sell the chip into.   For example right now there is big demand in the hobby CNC world for motor controller and chips optimized for CNC control.    A stepper controller though require a lot of domain knowledge and possibly specialized processes.   Another possibility is in instrumentation controller, multi function IC's to drive DMM, signal generators and the like.    As for imagining IC's there is a market but I would have to say your company would need to show a commitment to the chip.

The problem with imagining IC's is that you either deliver a complete solution or hand it off to somebody that can deliver that solution.   That is cameras are usually modules of one form or another.   Even in the hobby market there is significantly stronger interest in modules that deliver functionality.

So I don't think the idea is bad, especially if the device could come with public "how did we do this materials".   I'm big on technical education to put this in perspective.   I'm just not convinced that this is easy to do with an imagining sensor and in any event would test the waters with simpler hardware.   So while it might make sense for the right item targeted at the right market you will need to research what that is.

[up8051]

My proposition:
R2R Audio DAC (like old TDA1541, PCM63, AD1860 ...)

[Bassman59]

I'm leaning towards doing something like a 100% open-source design image sensor design intended for use with Arduino or very small processors. With a simple friendly readout and some moderately good specs. It would be to help get more of our research out into the real world where people could take advantage of it.

So, define "moderately good specs."

What's the pixel size? Full-well depth? Read noise? Number of rows and columns? Number of outputs? On-board digitizers or are the outputs analog? On-chip sequencer or external? Global shutter or rolling shutter? SDR or HDR? On-chip CDS or off-chip? And, of course, readout speed/frame rate?

[Berni]

The prices seam pretty reasonable but still rather high for a hobby project.

I have been instead experimenting with an idea of turning digital ASIC design into a puzzle game. Inspiration coming from the kind of games developed by Zatchtronics ( http://www.zachtronics.com/ ). This sort of design engineering based puzzle game genre is quite niche, but does have a decent flowing of people.

The game idea being a branching series of challenges to implement various digital functions by creating transistors on a silicon die. Starting with basic logic gates, building logic functions out of them, building an adder out of them, building an ALU out of them... eventually getting to things like a simple CPU. Trough the help of step by step design, easy block reuse, intuitive documentation, visual debugging etc hopefully a player with no digital knowledge can get trough it.

So far it has just been something i messed around with over a few weekends and came up with a rough prototype that can execute a inverter ring oscillator at up to about 1MHz in real time while rendering the chip structure in 3D.

[Haenk]

My proposition:
R2R Audio DAC (like old TDA1541, PCM63, AD1860 ...)

I would like that, but that's not going to happen. The parallel DACs need laser trimming (and then they went into binning them for normal and higher grades); that's the reason the industry came up with the cheaper bitstream DACs.
There are very good DACs available for little money, I doubt it's possible to DIY a better design without 30+ years of experience in that area of chip design.