Can you use 3d printers to print ICs and PCBs economically?

[Flux]

Can you use 3d printers to print ICs and PCBs economically?

Why use expensive fabs and expensive lithography masks when you can just configure a 3d printer to print millions of chips with less of a margin error?

Right now they are not dense enough(slightly less than 100nm) nor fast enough to compete with conventional fabrication techniques.

 

[3dilettante]

Can you use 3d printers to print ICs and PCBs economically?

Unlikely relative to manufacturers that make billions of these things a quarter.

Why use expensive fabs and expensive lithography masks when you can just configure a 3d printer to print millions of chips with less of a margin error?

What's the time frame here?
In order to print a million chips in a month, the printer would need to spit out the equivalent of a die every 2-3 seconds.

And how would this be less prone to error?

Right now they are not dense enough(slightly less than 100nm) nor fast enough to compete with conventional fabrication techniques.

You seem to have answered things at least in part.
What are the chemical and electrical properties of the filament materials?
If the base crystal is printed as well, well how can that be done piecemeal and still create a regular lattice?
Knowing that much of the point of silicon VLSI is the extremely uniform wafer crystal and tolerances measured in angstroms, how small is the physical variability of a 3D-printed IC to say six standard deviations? How much does its chemical concentrations vary?


If you want to make a toy project or educational exemplar with nowhere near the performance or quality of a regular IC, it could be done cheaply.
I don't know what it would offer over a cheap microcontroller or FPGA. There are garbage chips out there that would be better.

 

[Grall]

I really doubt you could print either PCBs or ICs. A PCB is a matrix composite of glass fibers in alternating directions/layers surrounded by epoxy resin, and I've no idea how the hell you'd print something like glass fibers. Don't think it's possible TBH. ICs - especially today at modern geometries - consist of super-pure silicon crystal, copper wiring, various oxides and something like several dozen different doping atoms. Again, how the hell would you print something like that?

I've seen examples of integrated circuits being printed with inkjet printers, but that was the equivalent of discrete components wired together on a sheet - like what electronics looked like mainly in the 1960s and 70s. Unless we're willing to go back to that era, a 3D printer revolution on the order of the evolution of the modern silicon integrated chip itself will be needed to go from 70s era circuitry to 21st millennium integration. All done with a mechanical system and considering the insane precision of the mechanics in a regular fab...not seeing that happening at any price less than many millions of dollars.

Heck, a person walking nearby the printer would be enough to fuck up the chip being printed - you'd need to build an entirely separate, vibration insulated house with climate control, moisture control, dust-free environments... Shit, smells like a chip fab! So just build a chip fab then, it'll be easier, simpler, very very likely cheaper, better and faster.

 

[tunafish]

Can you use 3d printers to print ICs and PCBs economically?

Why use expensive fabs and expensive lithography masks when you can just configure a 3d printer to print millions of chips with less of a margin error?

Right now they are not dense enough(slightly less than 100nm) nor fast enough to compete with conventional fabrication techniques.

I don't think you understand how fast a printer would have to be to compete with lithography.

A CPU or a GPU has billions of transistors. A lithography process builds all those transistors at the same time. A printer-style technology builds *one* thing at a time. So, to build a 1B transistor CPU in a reasonable time (let's say one minute), you'd have to build a single transistor in 60 nanoseconds. There is no way any printer will ever get anywhere near that.

 

[Lightman]

I don't think you understand how fast a printer would have to be to compete with lithography.

A CPU or a GPU has billions of transistors. A lithography process builds all those transistors at the same time. A printer-style technology builds *one* thing at a time. So, to build a 1B transistor CPU in a reasonable time (let's say one minute), you'd have to build a single transistor in 60 nanoseconds. There is no way any printer will ever get anywhere near that.

Not if you use printer 'head' with multiple nozzles. Then you would be able to print multiple elements at the same time cutting your estimate by a bit.
Anyway this is rhetorical as I don't know of any 3D printer capable of printing um structures at speed.
Even worse if you want to use hard material like titanium or alloys to print, as printing just one titanium hip replacement takes few months with current machinery.

 

[Grall]

Multiple print heads probably wouldn't work for the same reason reading or writing from multiple heads of a harddrive at the same time doesn't work. Minute differences in local temperature causes microscopic deformations in the geometry of the device (read/write head assembly in the case of a harddrive), causing heads other than the currently tracking head to go out of alignment.

Same thing would happen to a nanometer-scale integrated circuit being printed methinks. Not to mention, issues with vibrations with multiple independently moving parts in the same machine...

 

[3dilettante]

Can it be clarified just what sort of chip is being manufactured and the time frame?
This is being touted as a viable alternative to expensive fabs, right?

Is it too unfair to compare how this would build something like, say a Tahiti (edit: Hawaii) GPU?

300mm wafers = 150^2*3.14 mm2 per wafer x .85 (conservative wafer wastage number) / 438 (mm2 die size)

x the wafer starts per month of a fab (variable, can readily go above 100,000)

~= 13,700,000 dies manufactured in a month

30 days x 24 hr x 60 min x 60 sec = 2,592,000 seconds to do this

Dividing the throughput by the time gives about 5.3 die per second.
The GPU has 6.2 billion transistors, so 32.8 billion transistors per second.

How many nozzles are we talking about here, how fast are they, and what are the physical dimensions of them and their associated apparatus?
By the way, are we sure we need a real 3D printer here? One dimension is very underutilized compared to the rest.

 

[pcchen]

I think the 3D printing thing is mostly for making heavily customized parts. For something that's mass manufactured, 3D printing is unlikely to be competitive ever.

PCB is, of course, already "printed" (hence the name, printed circuit board). There were some research on printing a board "positively" (compared to current "negative" process), but that's mainly to simplify the process of making custom boards (imagine that you draw a PCB with some metal pens on a plastic board). However, they are unlikely to be economical for larger scale.

Chips are even harder to make with printing technology. Current chips have billions of transistors in just one chip. A wafer contains hundreds times more. No matter how fast you can "print" these transistors, it's going to take a very long time to even print just one wafer. A large fab is able to produce tens of thousand such wafers every month. It's even not very viable to print chips for custom purposes, and you can always use FPGA.

 

[karlotta]

kind of a 3D printer basic info;

http://www.fastcolabs.com/3016490/9-consumer-3-d-printers-for-every-budget

 

[Davros]

A better question would be
Can you use a 3d printer to print a 3d printer

 

[Blazkowicz]

Aren't chips printed already? Lithography is a kind of printing.
Next to that, 3D printers should be called 3D writers, i.e. it would be like replacing a monk who handwrites the book painstakingly with a robot monk, instead of using a printing press.

To continue the analogy, a problem of chip making is the mask is bloody expensive and can only print one thing. If you want to print a book, and prepare a whole special metal plate or wood plate for each page, that is gonna be expensive. But then we invented mobile characters, which allows characters to be reused instead of carving hundreds of thousands of them in one-use plates.

I have a hard time figuring out how that can be applied to ICs and/or 3D printers

 

[Blazkowicz]

I really doubt you could print either PCBs or ICs. A PCB is a matrix composite of glass fibers in alternating directions/layers surrounded by epoxy resin, and I've no idea how the hell you'd print something like glass fibers. Don't think it's possible TBH. ICs - especially today at modern geometries - consist of super-pure silicon crystal, copper wiring, various oxides and something like several dozen different doping atoms. Again, how the hell would you print something like that?

I've seen examples of integrated circuits being printed with inkjet printers, but that was the equivalent of discrete components wired together on a sheet - like what electronics looked like mainly in the 1960s and 70s. Unless we're willing to go back to that era, a 3D printer revolution on the order of the evolution of the modern silicon integrated chip itself will be needed to go from 70s era circuitry to 21st millennium integration. All done with a mechanical system and considering the insane precision of the mechanics in a regular fab...not seeing that happening at any price less than many millions of dollars.

(...)

I was going to reply "but you can print chip-like stuff with an inkjet printer!", then realized you addressed this.
Apparently there's progress on that front that I was not aware of, revealed in 2011. They printed a "CPU" and even made a 64bit DRAM.
http://www.technologyreview.com/news/423410/the-first-plastic-computer-processor/
Of course you need to have low expectations. Applications (if it works reliably at all, and with a scaled up design) would be in systems where you don't need performance - at all - and need complexity that is slightly higher than trivial, replacing a mechanical or electro-mechanical system, or discrete logic gates, or a 1MHz 8bit CPU with 128 bytes of integrated memory that sits on its ass doing nothing.

/edit : to keep things absolutely clear it's a DRAM made of 64 bits , i.e. 8 bytes of memory. It's also not compatible with the CPU.