Parts from same wafer

[Bruce4192]

I'm asking for help on any possible way to locate parts from same *wafer*.

 Can be bjt, mos, jfet or vacuum tube cathode material as its for physics experiments. Even op-amps possibly...
I've purchased large sealed lots of discretes for this purpose and now trying to sort them out.
I know it's a crazy sounding question.
Thanks.

[ataradov]

There is no chance of you identifying the same wafer. If the parts are from the same lot, they are likely to be from the same wafer, but not guaranteed.

And there is often variability within the same wafer, so if you need matching parts, you need to match performance characteristics of interest, not the wafer itself.

[Benta]

The closest you'll get is "cut wafer on sticky tape." But as atadarov says, even on the same wafer there are variations due to mask alignment etc.

[Bruce4192]

Yamaybe impossible to know for sure. I'm not looking for any improvement in matching, just devices 'born' at the same time like twins. Same crystal pull.

[SiliconWizard]

I don't know how you could do that. Certainly not just with electrical measurements.
Don't know what kind of physics experiments you have in mind. Are you going to extract the die from each part you got? If so, you could look for visual hints using a microscope. Maybe also chemically analyze the composition of pieces of wafer. Sounds like very advanced forensics.

[Oldtestgear]

Discrete used to have significant variation across one wafer, let alone across a wafer lot. From personal experience the variation can be very significant for small signal JFET devices & a wafer could produce up to 7 or 8 different part numbers.  In one case where matched individual  die were in a package, the whole wafer was probed & a print out made. Each individual die was marked with a serial number, the wafer was sawn & then an operator had the task of picking  the matched die by the serial number & putting in a waffle tray for assembly. These parts were always expensive due to the amount of manual labour involved in production.
Once packaged then the individual lot numbers would be difficult to trace through the system but NOT impossible for the OEM.
 
As you are in the USA, & assuming that this traceability is essential, you could contact Micross Components & ask if they would package die from a single wafer as a special order. There may be other companies that are still in business doing the same work but it is a very long time since I was involved in this industry. Note this is not a recommendation for Micross only that this company appears to be the largest left.

These comments apply to a much lesser extent to linear IC wafers as yield is usually optimised more easily & even less to digital parts.
Just by way of an explanation I worked in the semiconductor industry for PMI, GE/RCA/Intersl & LTC plus Eltek Semiconductors (UK die & wafer processor) for over 20 years.

FWIW Phil

[nctnico]

I recall some of the Agilent (now Broadcom) dual diodes where guaranteed to be next to eachother on the same wafer. But as others already noted: even those will have differences between them. If you want matching parts, you'll need to measure / sort all the parts.

[Bruce4192]

I recall some of the Agilent (now Broadcom) dual diodes where guaranteed to be next to eachother on the same wafer. But as others already noted: even those will have differences between them. If you want matching parts, you'll need to measure / sort all the parts.

I wonder why a customer would even want something like this. As mentioned, there is no improvement in matching etc.
Obviously, the substrate(s) are part of some external circuit in some form.

[Bruce4192]

I don't know how you could do that. Certainly not just with electrical measurements.
Don't know what kind of physics experiments you have in mind. Are you going to extract the die from each part you got? If so, you could look for visual hints using a microscope. Maybe also chemically analyze the composition of pieces of wafer. Sounds like very advanced forensics.

Sounds like insane physics but I've had success in long range, long duration non-local links using liquids. My goal has always been signaling. Solid state electronic components are convenient for test targets but a common birth and then treatment (laser,em and other) prior to experiment.

[daqq]

Not sure if useful, but STM32 devices do feature a user readable unique ID that does include information about the wafer number, position on the wafer and some misc info. See: https://pcbartists.com/firmware/stm32-firmware/generating-32-bit-stm32-unique-id/

Some do feature op amps, references or some analog parts.

[RoGeorge]

There are non-patterned Si wafers available on ebay, or even wafers with uncut chips on them, guaranteed same crystal.  Or you can order your own custom chip design.  There are plenty of fabless chips (e.g. Propeller microcontrollers), some companies are just consolidating orders for prototype chips (in very small quantities, e.g. for students, so the chips will be from the same wafer).  Another option is to use the in-house chip manufacturing capabilities of some universities (don't know which ones have in-house fabs, for example the University of Twente might have some in-house fab).

It is not clear what properties are you looking for, what is the physics experiment about?

[nctnico]

I recall some of the Agilent (now Broadcom) dual diodes where guaranteed to be next to eachother on the same wafer. But as others already noted: even those will have differences between them. If you want matching parts, you'll need to measure / sort all the parts.

I wonder why a customer would even want something like this. As mentioned, there is no improvement in matching etc.

Well, there is a some improvement in matching. With small parts like diodes, the error between adjacent devices should be relatively small.

[Bruce4192]

The series of experiments are trial and error attempts at non-local signaling. Ie two systems each in a faraday cage able to communicate.
It is not practical to fab on a hunch what topology the wafer should have.
There exists some guidance on arxiv but nothing hard.

The known working model came from this 2006 paper. But now moving on to electronic parts.

[ataradov]

LOL. Ok, we can move on. This is some BS published in a magazine created to enable "academic freedom" (read freaks that can't get published in real magazines).

Of course this voodoo requires matched silicon. As a next step look for ingots produced from the same beach sand.

[Bruce4192]

Yes moving on for sure.

[langwadt]

LOL. Ok, we can move on. This is some BS published in a magazine created to enable "academic freedom" (read freaks that can't get published in real magazines).

Of course this voodoo requires matched silicon. As a next step look for ingots produced from the same beach sand.

polished by naked virgins during a full moon

[EPAIII]

"Same crystal pull."

I think I figured this out at that point.

Yamaybe impossible to know for sure. I'm not looking for any improvement in matching, just devices 'born' at the same time like twins. Same crystal pull.

[RoGeorge]

The known working model came from this 2006 paper.

The paper is interesting, assuming the observations are correct.  Be very, very careful, particularly when human feedback is involved.  Placebo effects can mislead even the best intended researchers.  Never try to prove your own theories, always try to disprove them, because the easiest person to trick is oneself.

My first thoughts after reading the paper:
- the study was conducted on a very small sample, 4 persons, 2 of them being the authors + 2 their parents, which means the subjects were aware of the experiment, and aware about some expected outcome
- entanglement is synchronization between oscillations/oscillators (no matter if the oscillations are treated as waves or as particles).  Synchronicity does not last forever.  Once separated, the entangled oscillators will lose their synchronicity.  Decoherence is a big problem (e.g. in quantum computing the typical fix for decoherence is, firstly, cryogenic temperatures - less temperature means less noise - and on top of that, applying all kind of error correction techniques, though there is no visible progress in quantum computing, so it is highly unlikely to overcome decoherence).  Also decoherence happens almost instantly (in milliseconds or less, IIRC the max distance for entangled photons transported through a fiber optic, at room temperature, was in the range of a hundred km or so, so not exactly "to the other side of the universe" as it is told in popular science videos)
- in my opinion the conclusion from Bell's theorem are wrong.  Entanglement, as in synchronized particles or waves, do exist, but there's no spooky action at a distance.  The Bell's inequality appears because of all the random noises.  Doesn't matter there was a Nobel prize for that, the bell shaped curve in the delayed choice experiment, and the Bell's inequality is not a proof of non-locality, it's only the result of noises in the experiment.  However, I'm not a physicist, only looking into physics as a hobby, so I might have got it all wrong.  Time will tell.

All these doesn't mean you should stop experimenting with what you are after.
Get yourself some wafers, and best luck with the next phase of the research. 

[Bruce4192]

Maybe taking the Fairchild pjfets (4,000 qty) and using an oscillator on one sample. Then using the duts to influence the oscillator fft. This is something that would allow testing a large qty of parts. Temperature at 32f ice bath.

The spin coherence decoherence chapter of this book is also interesting.

[berke]

The paper is interesting, assuming the observations are correct.

The observations might be correct but their interpretation implies the following.

1) Heat soup in microwave
2) Eat half of soup
3) Throw the entangled other half in the garbage
4) Dump rat poison in garbage over the soup

You're now experiencing symptoms of poisoning.

You'd think that such an effect would have already been discovered after some billion person-years of human microwave oven experience.

[Bruce4192]

The paper is interesting, assuming the observations are correct.

The observations might be correct but their interpretation implies the following.

1) Heat soup in microwave
2) Eat half of soup
3) Throw the entangled other half in the garbage
4) Dump rat poison in garbage over the soup

You're now experiencing symptoms of poisoning.

You'd think that such an effect would have already been discovered after some billion person-years of human microwave oven experience.

Yes this remains a valid concern.

[Wolfgang]

The only idea that comes to my mind are military part, with JANS specs. Those must be tracable to a wafer, AFAIK.

[Bruce4192]

Just for completeness and full disclosure, I researched the metaphysics-Christian view of this spooky action-at-a-distance. My Catholic friend sent a copy of the Aquanis flowchart. It gives a green light to proceed.

[hans]

I think your best bet is to do manual matching of many parts. In audio space, there is some talk on matched transistors for audio amplifiers. People were very enthusiastic about JFETs such as the JFE2140. They are presumably made closely together on the same wafer, and then packaged together. But there are also other JFETs like that from boutique vendors like InterFET and LinearSystems.

Many opamps will probably also include similar structures for their input stage(s), as otherwise we wouldn't have 0.1mV-offset amplifiers without chopping techniques. However, designing for such tolerances is still a fine art in fabrication techniques and resilient structures. Absolute tolerances on silicon is often piss poor (+/-20% not uncommon). Relative tolerances can be better.. but are still subject to how the properties of the material (e.g. doping etc.) progress across the material. I don't know how uniform or predictable these changes are.. e.g. whether its possible to model them as increasing in X/Y axis (or any argument of that vector), and then creating IC structures that can cancel out imbalances of all these randomly distributed directions.
Its the kind of layouting stuff thats not my cup of tea.. I've only heard some stories about it during my time at uni doing EE courses. Its the time I decided its fun to (sort of ) know but not for me to do.

[berke]

How about slicing a BPW34 photodiode in half?  They're fairly big (about 5 mm by 5 mm) and symmerical so I think you would get two working photodiodes.