A relatively cheap way of offering low-volume custom "chips"?
markaeric
Posts: 282
It seems that a lot of people would love to have a SoC tailored to their needs. If the expected volume is low, then it is understandably just a pipe dream. However, what if there actually was a viable option? I think there just might be, and it's sort of based on the concept of stamps - where numerous components are placed on one "chip". One of the biggest issue seems to be about footprint, and costs. Both of these can at least be slightly reduced utilizing bare dies instead of packaged ICs. It's pretty common for cheap electronics to utilize IC dies placed directly on circuit boards by wire bonding the IC pads to board pads, or even simpler, by using flip chips. This is even common on all major microprocessors. so what happens if you take this concept and apply it to highly populated micro circuit boards such as stamps?
As an example, what if you could integrate a processor die, ram die, ADC die, and all necessary support components such as voltage regulators and crystals on a smallish board? At the same time offer an option with flash instead of ram, and a DAC instead of ADC. You could do this economically if your chips consist of individual components rather than having them all on a monolithic IC. You spare yourself from the risk of creating an entire IC with these features and not getting a return on your investment, because you'll have cheaper and simpler components that can be used in one configuration or another, and built basically on demand. Such an option would be particularly good for people who would like to incorporate advanced features and functionality in a small run of commercial products without having to worry about complex multi-layer board design. There are computer-on-board options, but as far as I know, none of them are customizable (at least cost effectively), and they're certainly not very cheap. Of course, this doesn't have to be limited to computer-on-boards, but simpler and flexible microcontrollers. I could be wrong, but I think Parallax is in an excellent position to offer such a product - big and capable enough to produce custom ICs, and small enough to cater to lower-volume customers. The Basic Stamp brought them to where they are today, why not take it to the next level?
I'd like to hear some of your thoughts on this.
As an example, what if you could integrate a processor die, ram die, ADC die, and all necessary support components such as voltage regulators and crystals on a smallish board? At the same time offer an option with flash instead of ram, and a DAC instead of ADC. You could do this economically if your chips consist of individual components rather than having them all on a monolithic IC. You spare yourself from the risk of creating an entire IC with these features and not getting a return on your investment, because you'll have cheaper and simpler components that can be used in one configuration or another, and built basically on demand. Such an option would be particularly good for people who would like to incorporate advanced features and functionality in a small run of commercial products without having to worry about complex multi-layer board design. There are computer-on-board options, but as far as I know, none of them are customizable (at least cost effectively), and they're certainly not very cheap. Of course, this doesn't have to be limited to computer-on-boards, but simpler and flexible microcontrollers. I could be wrong, but I think Parallax is in an excellent position to offer such a product - big and capable enough to produce custom ICs, and small enough to cater to lower-volume customers. The Basic Stamp brought them to where they are today, why not take it to the next level?
I'd like to hear some of your thoughts on this.
Comments
There are also FGPA's with analog sub-systems. See Actel's products.
Also Parallax doesn't fab IC's. You still pay $$$ for a custom chip no matter what. You still have to design, prototype and test any ic yourself. The tools for this are very, very expensive.
The ceramic substrate usually contained traces to connect the chips together as well, so it really amounted to a miniature pc board made from ceramic instead of fiberglass. They were smaller, conducted heat better, and usually provided some speed improvement, but had their own set of limitations.
This is extremely common on extremely cheap electronics such as calculators, and in fact, this method is used for the main controller in a $10 wireless mouse I have disassembled right in front of me. This is definitely cost effective for complete end-user products, such as the devices I mentioned, as well as for "chips" that have features such as large amounts of ram which would require huge die sizes and in turn, be more likely to have defects. Another benefit is that it would allow the chips to be comprised of ICs made under different processes, so all the sudden you could have something like a propeller with, for all intents and purposes, "on-board" flash. Of course, this probably isn't the cheapest route to go if you intend to sell millions and millions of a particular chip, but that's not really the target I had in mind. It's really more about taking the concept of the Stamp to the next level, while eliminating the wasted space and cost associated with the packaging for each individual IC.
Now for certain unique requirements like the sort the military always have and have a truck load of money to throw at exotic solutions, it might make sense.
Yes, chip on board is very popular for high volume inexpensive devices and has been used for a very long time. The down side of the technology is the reduced life span of the products. Not a big deal for products like mice, calculators, etc. that usually fail for other reasons.
@rod1963
Monolithic circuits were used by the military and NASA quite a while before the 80's and the attempt at wafer scale integration. I don't think ceramic substrates will ever compete with high volume SOC's or MPU's but advances in manufacturing like 3D printers could reduce the cost and increase their use.
From what I've read, it appears that CoB is *more* reliable due to reduced interconnects and stresses placed on solder joints by heavy packaging. Since circuit boards are hardly repaired these days - especially if they utilize a lot of BGA components - a faulty, easily removable chip would be cheaper to replace than an entire PCB, just like people can do with Basic Stamps.
That could very well be so when comparing CoB to SMT. I was comparing CoB to hermetically sealed ceramic monolithic circuits. The epoxy "blob" used to seal the chip on CoB boards does not provide as good a seal, so eventually moisture does get through and attacks the chip. This is not to imply that ceramic mono's are perfect. They have also been known to fail due to moisture, but much less frequently than CoB's.
Ah, yes. No need for all that fanciness, since it wouldn't be intended for use in F-22s, etc.
http://www.pulsonix.com/chiponboard.asp
I had a play with it when I was beta-testing the new release, and it's quite easy to use.
Exactly. I think CoB is a great way to go for many products. I would not think twice about purchasing a laptop, desktop, cell phone, DVD player, etc. that was manufactured using CoBs, but for high reliability systems like aircraft, implanted medical devices, etc. something more reliable is needed.
That`s a nice feature. I wonder if you could do the same thing with a package that does not have specific support for CoB`s by defining an appropriate sized package and pinout for the required chip.
http://www.ghielectronics.com/catalog/product/373
To some extent, but not exactly.
The wafers on ceramic approach will be expensive. First off you need a precision tool to place and connect the chips, not some $500 piece of plastic junk you buy at a Maker faire. Secondly you tooling jigs, the ability to test each wafer before place and routing them, then testing the whole unit. And it gets better, who will sell you wafers? You just can't goto Mouser, you talk direct to a Freescale or Nxp, I'm sure they'll sell you say 10k-100k of them to you, but nothing less. That means you have to have a market that is willing to buy or you end up bankrupt real fast since all your capital is tied up in wafers.
I am not suggesting that Chips on Ceramic substrates (hereafter referred to as CoC's) would replace or compete with CoB's or the tiny boards todays smt chips make possible. What I am suggesting is that a substantial reduction in price would make them competitive in a broader range of applications.
CoB's are the technology of choice for high volume low cost relatively simple consumer products. PCB's and standard packages (which are not that much larger than the chips they contain) are used for lower volume somewhat more complex products.
Low volume specialized electronics equipment covers a much wider range of environments. Equipment that operates in the room temperature range can use CoB and SMT on standard pcb's, and automotive applications are only slightly more demanding. Military, space, medical, and some industrial applications (oil drilling for example) are much more demanding. These are areas where CoC's may be required, and a reduction in cost would increase their use.
All my prototype stuff is ONLY prototype, and never needs to be high reliability. It just needs last long enough to prove the design. I seldom get invited to embed projects in patients. And my F-22 has other maintenance arrangements.
Suppose some crazy person wants to move this forward. What would you have us do next?
I'm not sure how many vendors sell bare dies, but I know at least some do. Of course, the ideal way is for a company to design all (or most of) their own IC circuits such as processor cores, ADCs/DACs, SERDES, OPAMPS, etc. to ensure simple interface compatibilities. Then design numerous different "chip board" layouts to make various different chips using combinations of components available. Taking it one step further, if a customer would like a customized design they could pay the manufacturer to develop a chip board out of the various components to their specifications - within the appropriate constraints, of course.
The most ideal system would allow a customer to dictate the specifications, and then an automated system design the chip board, with minimal human involvement, significantly reducing the costs of development, and being infinitely cheaper than putting it all on one custom die as in the case of ASICs.
Fiberglass PCBs are probably much cheaper, and definitely more common. If you want, you might even be able to eliminate most traces by doing die-to-die wire bonding, so they can be placed very close together. As far as I can tell, even high-end x86 CPU dies are placed on multi-layer non-ceramic PCBs, so it should be more than sufficient for the lowly devices I had in mind.
How to they do the wire bonding anyway?
Flip-chip placement would probably be easier than wire bonding. Either way, a reprap is unlikely to be accurate enough. Depending on the die sizes, you might be able to get away with wider than usual pitches per component since everything is not crammed all in one IC - at least for bigger components such as a processor.
I don't think this is a task you could perform in a garage with hobby-level tools, most definitely not the IC manufacturing. The most likely and cheapest way is to have it all made in (guess where).
It is becoming more common to get what looks like an IC on the outside, but is a multi-die package on the inside.
Raspberry PI has one, and there are devices like Freescale'S engine controllers MC33813/MC33814 in QFN48.
http://www10.edacafe.com/nbc/articles/1/1112248/Freescale-Introduces-Analog-ICs-Small-Engine-Electronic-Control-Units
or a similar thing from Infineon, who started with TLE78xx multi-chip module and now have TLE983x devices. (also in QFN48 package)
http://www.infineon.com/cms/en/corporate/press/news/releases/2012/INFATV201206-045.html
Nice parts, a Microcontroller plus higher voltage and power ...
I would like stuff that needs to be very close to the micro controller to be built in like a couple A/D converters. Could they be made software configurable to act as two at half precision, or one at full precision?
Also, I'd want all the standard stuff built right into the package if possible. crystal, regulator, pull ups. USB to serial converter. capacitors between power and ground.
I'd THINK I'd want the 9 axis inertial nav stuff: compass, accelerometer, and gyro; and barometric pressure sensor. Maybe GPS, too
I also think I would want a separate prop chip connected to each peripheral, and all the prop chips bussed together on either A) just two lines, or
Assuming I actual want all the stuff listed, which of it makes sense to include this way, and which does not (and why)?
Would the cost actually be better than buying and building separate components? At what volumes?
If the answers to these are good, I would be interested in looking into this further.
I don't remember where I saw it, but someone had uncapped some chip, just to find in one corner was the main IC, and in the other corner was a flash IC. Funny thing is, the company was touting the device's secure code capabilities, but the person who uncapped it accessed the "secure" code in the flash easily because it was a standard part, and it wasn't actually encrypted.
@prof_braino
Sound like you want a complete product in the smallest possible footprint. I suppose it's possible - and it leaves me wondering if interference would be more of an issue, less, or about the same.
When it comes to cost, I guess it depends on a lot of different factors. When using individualized ICs, you can take into account the lower price per die, and lower financial impact for faulty parts (a big deal if it is highly integrated, because one small defect can render the entire IC useless). Another benefit is that your chip can contain ICs made by different processes. The negative side is the added cost of packaging, though that might be offset (significantly) by the reduction of components needed in an end-product. Overall, there is probably a cost benefit or else why would some of the cheapest electronics be using CoB? Most likely, there is a fine line between the level of integration of individual ICs, and the manufacturing volume, I'm not going to pretend to know where that line is.
Speaking of multi-prop set-ups, here is something that I think would be cool: 2 (or more) prop dies without the 32k hub ram (still has the hub, though) then a separate die that is just ram to allow for higher overall capacity, that allows multiple ram dies to interface with each other through some shared memory locations. That way, you wouldn't have to waste a cog just for prop to prop comm.
I think the two biggest advantages can be seen when:
1) Level of integration is very high (such as large amounts of ram/rom on-chip
2) Cheaper than ASIC for lower volumes
Most of the inexpensive products I have seen ( I suffer from the "it was dead so I took it apart to see what was in it" syndrome. ) that use CoB's typically have only one or 2 CoB's and a few passive components on the board. AFAIK putting the bare chip in a package adds a hefty amount to the price of the chip (percentage wise) so bare chips are a great way to reduce the cost of a product if you have the volume to justify the added production equipment cost and persuade the manufacturer to sell them to you.
@prof_braino
What you want to do sounds more like a complete system module. It would probably end up being bigger than most or possibly all of the chip packages available today. It could probably be done using CoB technology in a package about the size of a credit card or a bit smaller if you can get bare chips with the functions you require. The ADC chjp you describe does not sound like anything that is currently available.
In low volumes this system would not justify going to CoB's, but could be built at a reasonable price on a somewhat larger board using smt components and a multi layer board with components on both sides.
I think this is mostly plausible for companies that design (not necessarily manufacture) their own IC components, so there's little to worry about when it comes to obtaining bare dies from some other company. The best and easiest components to source from some other company would be things like ram and flash. Also, why purchase your own production equipment when there are plenty of assembly houses that can do CoB? And lastly, I wasn't thinking these would have to be put in some kind of standard package - would that really be necessary?
Could be, this is not an area I am all that familiar with. Still, having custom chips manufactured involves considerable up front costs so it would require either high volume low cost or low volume high priced products to justify it.
True, but why limit yourself to just ram and flash if you can build your product with off the shelf chips. Having custom chips made is expensive. Why do it unless there is a cost saving or technical/maketing reason to do so.
Very good point. They may have enough clout with suppliers to make getting parts like bare chips possible. The only reason I could see for doing this in house would be if you could do it for a lower cost or if quality control was an issue.
No, not really necessary, but it might be desirable in some cases. If you wanted to build a faster more powerful replacement for an existing product using the same package and pinout would be necessary. If it was a new product like the one from prof_braino's wish list (post 27) it would be larger than any standard package, but it would be a good idea to make it easy to mount to a board with standard connectors or ribbon cables. Perhaps something along the lines of the GG boards. Then again, considering how much he wants to cram on the board perhaps a power/data/inter board communications bus on one edge of the board and I/O connectors on the other 3 edges.