wafer.space Community - 🏗️-die-sorter

@Andrew Wingate - This look okay?

1

Hey everyone, like Tim mentioned, we’re building a die-sorting robot! To give you a sense of scale, here’s what one run looks like:

50 full wafers
28 reticle shots (each a “group photo” of all projects) per wafer
40 unique projects per reticle shot
Add it up and you get 56,000 project dies
Each die is about the size of a few grains of rice: 3.88 mm × 5.08 mm
Laid out flat, that’s over 8 ft² (0.8 m²) of silicon — basically a coffee table made of chips In electronics manufacturing, tape-and-reel is the standard way to package components, so that’s our target. We looked at wafer packs (think plastic ice cube trays for chips) and gel packs, but reels make the most sense at this scale. The wafers will arrive pre-sliced and mounted on tape frames. With so many dies to handle, we’re going straight from wafer to reel — no detours. I can’t wait to see all the designs people come up with, and I’ll keep posting updates as this project rolls forward (pun intended).

(edited)

@Andrew Wingate - 3.88 mm x 5.08mm is a lot bigger than a fingernail clipping unless your not clipping your fingernails a normal amount

1

Yeah.. it's kinda close in sq units. it's a thumbnail at 1mm

2:21 a.m.

These are very early models, just exploring how all the pieces might fit together into a compact machine; after all, this whole setup has to ship to Singapore! The concept is straightforward: the wafer sits right in the center, surrounded by 40 tape-and-reel stations (20 on each side). Each station will load a specific die into its own reel and then seal it up with protective film, ready for shipping :)

1

https://docs.google.com/document/d/1dXzGoGS7ERWlQNyJ9aNEIOZe-R9AS1tskxvqsQ3FO48/edit?tab=t.0

10:47 p.m.

https://docs.google.com/spreadsheets/d/1coUm6J2RCxoX6v4eKYzD-owmuKActA8mDBkgD4J9ECg/edit?gid=196570729#gid=196570729

Resolution Conversions

2:10 a.m.

@Andrew Wingate - Did you also see https://foss-eda-tools.googlesource.com/third_party/shuttle/gf180mcu/mpw-000/foundry/+/refs/heads/main/manifest-gfmpw-0.csv ?

2:11 a.m.

@Andrew Wingate - @Leo Moser (mole99) is working on the tool which stitches the retcile together @ https://github.com/wafer-space/reticle-stitcher

GitHub - wafer-space/reticle-stitcher: Tool for stitcher an MPW int...

Tool for stitcher an MPW into a full reticle for delivery to a foundry. - wafer-space/reticle-stitcher

Sadly it doesn't seem like the GF180MCU/MPW-001 which are associated with the die I gave you was never published.

@Andrew Wingate - Reading the IDs - https://siliconpr0n.org/archive/doku.php?id=vendor:efabless#reading_efabless_id_numbers

@Tim 'mithro' Ansell https://github.com/evezor/waferspace_die_sorter/wiki/Overall-Strategies

Overall Strategies

Contribute to evezor/waferspace_die_sorter development by creating an account on GitHub.

@Andrew Wingate - On the holding of the wafer frame, I'm unclear if you need to do something to the wafer frame to allow easy removal of the die.

Thx for some of the info Tim, -edit for those watching, we're looking to put dies into smt tape EIA STANDARD - 8 mm Through 200 mm Embossed Carrier Taping and 8 mm & 12 mm Punched Carrier Taping of Surface Mount Components for Automatic Handling (edited)

EIA-481-2003.pdf

633.54 KB

9:06 a.m.

The process, packaging, and materials cannot impact the components. This includes their markings and their electrical and mechanical characteristics. Additional guidelines exist for further reference, including: EIA-383—Preparation for the Delivery of Electrical and Electronic Components EIA-541—Packaging Material Standards for ESD Sensitive Items EIA-583—Packaging Material Standards for Moisture Sensitive Items EIA-624—Product Package Bar Code Label Standard for Non-Retail Applications

9:07 a.m.

I came across an embossing machine https://www.youtube.com/watch?v=4-XQ0Mk-yMQ

BoardWin Technology

Embossed Carrier Tape Forming Machine with Vision System + Slitting...

9:08 a.m.

9:12 a.m.

Hey Tim, I was just looking around a bit for some info on smt tape. Found a generic supplier and it seems like they have a min order of 5km https://www.alibaba.com/product-detail/Plastic-Packaging-Materials-for-Electronic-Components_1600512330442.html?spm=a2700.prosearch.normal_offer.d_title.186d67afspjYlH&priceId=1b90f94702bb4f568b4c91a9b0425c83 They also have tape, which seems to be around a dollar at qty 100. They never give the real prices until you start talking with them. https://www.alibaba.com/product-detail/13-3-MM-Width-Smt-Cover_1600521271655.html?spm=a2700.details.you_may_like.2.4a234424fGJnNX

Plastic Packaging Materials For Electronic Components Packing Ps Ca...

Plastic Packaging Materials For Electronic Components Packing Ps Carrier Tape - Buy Packaging Materials packaging Materials For Electronic ps Carrier Tape Product on Alibaba.com

13.3 Mm Width Smt Cover Tape - Buy Cover Tape narrow Width Tape pac...

13.3 Mm Width Smt Cover Tape - Buy Cover Tape narrow Width Tape packing Tape Product on Alibaba.com

@Tim 'mithro' Ansell I've been playing with a few cameras and microscopes that I have. That 2mp camera I was talking about cannot read these IDs. Also the crappy usb microscope I have also does nothing to help. I also have a 48MP microscope and it also cannot read these id's. What I have shown here is about the best zoom I'm able to get out of it. It's possible with some better light we could get a better look, but at the end of the day these are very very small features.

I would have tought you'd use the position on the wafer to sort, not trying to read those tiny IDs.

I guess you could also visuallly compare the pattern with a reference, if you want to go optical

I came up with this and wanted your opinion of if this is a viable option. The idea is to create larger characters that can fit in every corner. The process and run could somehow be encoded if desired, but all that info is in the id anyways.

tnt

I would have tought you'd use the position on the wafer to sort, not trying to read those tiny IDs.

Tim has expressed interest in verification steps along the way. Additionally, there seems to be little trust that all wafers will be the exact same (even if this is just how things are mounted on the wafer frame after dicing.

How many projects are you going to have on a wafer?

urish

I guess you could also visuallly compare the pattern with a reference, if you want to go optical

That's also a good idea, but not sure how easy it would be to establish some kind of baseline

urish

How many projects are you going to have on a wafer?

40 projects per reticle shot 28 full reticle shots per wafer

9:30 a.m.

There will be some partials at the edges and sides that will be discarded afaik

So we need less than 6 bits of data. I don't think letters are the best way to attack this (but they may just be good enough)

Originally I had thought just naughts and crosses and 'None' could do it, then thought why stop there

9:32 a.m.

alphabet alone is 26^4 = 456,976

The other question is whether you also want to encode the run id somehow

9:33 a.m.

in case one wanted to identify which wafer the die belongs to

There is also the process and the run, yes

9:33 a.m.

There is always still the ID

but is that for the #🏗️-die-sorter

I am not suggesting we remove that

urish

but is that for the #🏗️-die-sorter

Correct, basically just the die sorter

so I'd just keep it as simple as possible. Maybe like 2x2 matrices on each edge, that gives you a total of 16 bits, so you could actually encode the data twice + parity? You can also use some of the bits for detecting the orientation

9:37 a.m.

Looking at the space, triangle shape (3-bits) will probably fit better

urish

so I'd just keep it as simple as possible. Maybe like 2x2 matrices on each edge, that gives you a total of 16 bits, so you could actually encode the data twice + parity? You can also use some of the bits for detecting the orientation

That's also a great idea.

something like:

9:38 a.m.

1

9:39 a.m.

then you can just cut these shapes out and start playing tetris with the pieces

1

9:41 a.m.

looking at the samples you send, the padring has one darker edge - is that always the case? If yes, then you can use that for orientation

9:41 a.m.

(in any case orientation needs to be sorted out for packaging)

No, I believe that is just an artifact of my bad lighting

1

9:47 a.m.

Above, Tim had shared an example output of the reticle stitcher https://discord.com/channels/1361349522684510449/1419841819138854993/1420231095215067169 So once a couple of dies are found, creating the map for picking should be prett straightforward after that

1

@Leo Moser (mole99) - Thoughts on IDs?

5:01 p.m.

@Andrew Wingate - Do you have pictures of your microscope / camera set up?

BTW, are raw dies on reel something common ? (i.e. would packaging house usually be equipped to handle that)

5:05 p.m.

Side note: I tried reading the die-ID on my MPW1 chip with my microscope and no way ... I can barely make-out the OSHW logo and that's if I light it at just the right angle ... (edited)

Guess I got too use to having access to @digshadow's higher end equipement

1

5:30 p.m.

@Andrew Wingate / @Leo Moser (mole99) - We probably want a "pin 1" type marker as well for orientation.

@Tim 'mithro' Ansell how quickly do you need something? maybe sync up this weekend?

@digshadow - Well, having an ID which is readable by people beyond your friends seems like something we should do

Hmmm? Need more context

tnt

BTW, are raw dies on reel something common ? (i.e. would packaging house usually be equipped to handle that)

Dunno - but reels are the preferred format for PCBA houses normally.

Yeah, because of pick and place machines, but if the machine dealing with dies aren't made for those, that won't help

digshadow

Hmmm? Need more context

At the moment the IDs which where used on the Google shuttles don't seem to be easily readable by the equipment that @tnt or @Andrew Wingate have. Thus it might be good for us to go with an ID which is bigger and allows more people to read them easily without having to get a good microscope from you

1

tnt

Yeah, because of pick and place machines, but if the machine dealing with dies aren't made for those, that won't help

@tnt - Yeah, these are the type of questions we are trying to sort out now, I'd like to do at least some type of short PCBA with wire bonding run (probably using GFMPW-0 or GFMPW-1 ICs) ASAP.

Tim 'mithro' Ansell

@Andrew Wingate - Do you have pictures of your microscope / camera set up?

Pardon the mess. This is the microscope I used to send the picture above. The crappy little black usb microscope on it's side also did not seem to work very well. Some of it could be bad lighting, user error, or they really are just too small to be read by crappy equipment.

1

Tim 'mithro' Ansell

@Andrew Wingate / @Leo Moser (mole99) - We probably want a "pin 1" type marker as well for orientation.

Yes, a pin1 is probably highly useful

2:04 a.m.

I was trying to come up with something that would still be at least partially useful if the dies were cut up further, but I think now that's another problem.

Expanding on @urish idea of some bitwise information encoded into the corners we could also go maybe as far as something like this if the corners of the triangles are always black--or whatever color--that could be used to identify location/skew/etc. and if triangle has a side length of 5 that leaves 12bits of info per corner--this obviously does not count for the pin1 corner though. The deeper we dive into this, the more it seems like it may be useful to keep these human readable.

@Andrew Wingate - I think @Leo Moser (mole99) thoughts are important, as he is the one who will have to implement it

Tim 'mithro' Ansell

@Andrew Wingate - I think @Leo Moser (mole99) thoughts are important, as he is the one who will have to implement it

Oh, of course! I was just throwing anything at the wall to see if it sticks. You are correct, I've been thinking more application than implementation. @Leo Moser (mole99) I would love some advice on what's possible and easiest to implement.

2:29 a.m.

@Leo Moser (mole99) I see on this chip I have there are ?different layers? that have different colors. Is it possible to do something like what's pictured here for the best contrast?

2:32 a.m.

It seems there are a couple dies on this wafer that have done something similar

Maybe people in the SiliconPrawn discord might have thoughts - they have spent lots of time looking at various markings / doodles on chips.

IME contrast / color is extremely dependent on how you light the samples.

I'm on my way back from Austrochip, so I can reply to questions again :) I started gf180mcu-precheck a few days ago. It is a custom LibreLane flow to read the user GDS, perform various checks (e.g. top-level, density, etc.), insert the ID, and whatever else we need. At the end, you receive the final GDS used for fabrication. I use the KLayout API to generate the shapes for the ID. Currently, I'm using the built-in TEXT PCell to generate the digits for the ID, but we can use anything else: a QR code, a pattern of dots etc. One thing to consider: we should not specify a fixed position for the ID. Instead, we provide an abstract cell for the ID that the user can place anywhere on their die, which is later replaced with the generated ID. This allows to do custom padframes (or no padframes at all). I can extract the position of the ID cell during the precheck. With the location of the die on the reticle and the offset of the ID, you can position the camera above the ID. As for the layers, I would place the ID at Metal5 (topmost layer) with a no-fill shape around it. I assume this will give enough contrast. There are ways to achieve different colors with diffraction patterns, however, I'm no expert on that :)

1

We did some experimentation with diffraction patterns on ttihp0p3, we may have some results when we get back the chips (Dec?)

@Leo Moser (mole99) / @Andrew Wingate - I responded at https://github.com/wafer-space/reticle-stitcher/issues/1#issuecomment-3346714007

Die ID location and design · Issue #1 · wafer-space/reticle-stitcher

Hey @mole99, I figured this was the best place to continue this conversation about the ID location, design, and layer. You posted: One thing to consider: we should not specify a fixed position for ...

4:05 a.m.

I would love people like @tnt to be able to identify they got the right part and the pin zero / orientation

Tim 'mithro' Ansell

@Leo Moser (mole99) / @Andrew Wingate - I responded at https://github.com/wafer-space/reticle-stitcher/issues/1#issuecomment-3346714007

@Tim 'mithro' Ansell and @Leo Moser (mole99) For reference, the first pic was taken with my phone. Pretty new samsung. Second is taken with one of those super cheap usb microscopes

Andrew Wingate

@Tim 'mithro' Ansell and @Leo Moser (mole99) For reference, the first pic was taken with my phone. Pretty new samsung. Second is taken with one of those super cheap usb microscopes

The cheap microscope looks quite good already! Did you take the picture with your phone in macro mode?