wafer.space Community - Info from Disco Engineers

Like I said above, went to Disco today and they were really great people and offered me a lot of time. They also seemed very willing to give us the benefit of doubt and work with us for our special needs.

10:06 a.m.

First up, @Leo Moser (mole99)

Saw Streets and Back Grinding

For full thickness they wanted 100um saw street. And then they also wanted a callout for backside chipping. They also said 70um for the background would be more appropriate.

10:07 a.m.

10:08 a.m.

Also @leo, they showed me a picture and the saw street was <50um not sure where that was coming from

10:08 a.m.

10:11 a.m.

Test features and sawstreet.

So in our reticle design we had 30um on each side to give us the total 60um saw street. GF did not honor another 30um and started their test features right there.

10:15 a.m.

Dicing Metal

We do not want to dice through features or test patterns. Apparently it loads up the wheel and that causes heat/ time from wheel dressing and other bad things. First one, they requested we ask GF to not let test features cross our dicing lines. Second one (and I don't even know how/if this is even possible?? but they asked if there was a way for GF to not have metal deposited in the un-shot spaces

In the pic posted with < 50um dimension, AFAICT that's just the polyimide deposition. but you can cut through that, doesn't matter for anything, there is nothing there. The 60.90 is the actual point where things start to matter ( that's the Metal5 from the guard ring which extends all the way to the edge of the project dies ) (edited)

1

we have that little sliver, that was causing all kinds of problems. They asked that we have no features that are taller than they are wide. No sliver there would be best. but that's something we already knew. I asked for the pictures they showed me a bunch, I will share them when I do. You could plainly see the one good cut and then the second cut came along and it was all crap.

10:25 a.m.

Dicing plan

They really don't need a whole lot it turns out. We need to give them the centerlines of every cut. They load those in and it just patterns after that. Basically said the table like we have and a picture of the reticle helps.

Main takeaway

We need to call out the part that is the test pattern and the little sliver if it exists as if they were dies and they should be a part of the table too.

Bonus stuff

They do laser dicing, they gave me a back of the envelope quote, the sawstreets can be super thin, but it is expensive. They do package dicing, I showed them the PCB and they were thinking it needed to be diced until I showed it was just v-grooved. -# That's all I can think of at the moment, I have to make a meeting in 10min. I think there's more to share later

So, let's finally take a look at this in more detail. @Andrew Wingate The reticle stitcher does not differentiate between a saw street and a callout. Everything between two user projects is designated as the saw street, so the only relevant factor for me is the total width required. Do I see it right that the initial suggestion from them was 100um saw street + 2150um callout? That would be 400um in total? That's huge... Even if we rotated the reticle by 90° and lost a few slots that way, Tim showed that we can only get to a 281um total saw street. Otherwise, we would have to remove a column of projects. I also see a 75um callout in your sketch. Would 100um + 275um also be an option they can work with? i.e. 250um in total?

Sorry away from computer so will try my best. The extra 150s were for backside chipping callout specifically. They showed me a number of different chippings and said that anything that shouldnt propogate or affect the die should be fine

9:47 a.m.

The 75um callout was a desired width with backgrinding. The 100 was full thickness

9:49 a.m.

I also said Januarys batch would be identicle to these and the same for run 3. They have seemed pretty agreeable so don't forsee any issues

9:49 a.m.

The different layout is a larger call. Not sure how to approach that one just yet.

Thanks for the reply. Sorry but your sketch is a bit hard to read. I only see 100um as the saw street width, not for the callout? Or did you mean to say 150um instead of 100um? So 75um callout with backgrinding, and 150um callout without backgrinding? And the saw street is always fixed at 100um. But then again, is backside chipping such a big deal to us? Also, to keep this in relation, we are now doing 60um total width with backgrinding, which is significantly smaller than what they would like and it still seems to work reliable for Run 1-3 on their end?

Leo Moser (mole99)

Thanks for the reply. Sorry but your sketch is a bit hard to read. I only see 100um as the saw street width, not for the callout? Or did you mean to say 150um instead of 100um? So 75um callout with backgrinding, and 150um callout without backgrinding? And the saw street is always fixed at 100um. But then again, is backside chipping such a big deal to us? Also, to keep this in relation, we are now doing 60um total width with backgrinding, which is significantly smaller than what they would like and it still seems to work reliable for Run 1-3 on their end?

saw street: 100um with no backgrinding 75um with backgrinding. Forget about the 150, that is an allowance for them to have bad things on the back side. Has nothing to do with saw street. They left town this morning, but I will be emailing them to send me all the pics I took, some of them were pretty informative. Hope that's more clear.

Andrew Wingate

saw street: 100um with no backgrinding 75um with backgrinding. Forget about the 150, that is an allowance for them to have bad things on the back side. Has nothing to do with saw street. They left town this morning, but I will be emailing them to send me all the pics I took, some of them were pretty informative. Hope that's more clear.

Alright, to be honest I'm still confused by the callout - but doesn't matter!

So, if I increase the current saw street width from 60um to 100um, everyone will be happy?

Leo Moser (mole99)

Alright, to be honest I'm still confused by the callout - but doesn't matter!

So, if I increase the current saw street width from 60um to 100um, everyone will be happy?

Yeah, if that's possible they should be happy.

Nice! Yes, we'll have to switch to a different reticle layout, which reduces the overall capacity by a few quarter slots, but other than that, it should be fine. I'll update the reticle stitcher in the next days to give this layout a try, so we can be sure that it fits, and we know what the new sizes for the quarter- and half-slots are going to be.

1

@Leo Moser (mole99) - Can we go for 200um instead? Better to have a saw street which is wider if possible? Could we just make the 1x1 slot (and half slots?) grow slightly bigger to keep the quarter slot the same size? That way you can always drop a past run's smaller design into the slightly bigger slot?

9:47 p.m.

@Leo Moser (mole99) - I think @Andrew Wingate was said something about Disc didn't wanting to dice through the GF added test structures? So we need the right amount of padding around the outside?

Tim 'mithro' Ansell

@Leo Moser (mole99) - Can we go for 200um instead? Better to have a saw street which is wider if possible? Could we just make the 1x1 slot (and half slots?) grow slightly bigger to keep the quarter slot the same size? That way you can always drop a past run's smaller design into the slightly bigger slot?

As I said, we can go up to ~250um total saw street width. However, the larger the saw street, the less area there is for the projects.

Could we just make the 1x1 slot (and half slots?) grow slightly bigger to keep the quarter slot the same size?

In theory, yes, but this will increase the required area for the reticle. I'm also not sure if we can keep the saw street width at 200um without removing a column of projects then. If you want me to, I can explore that option.

That way you can always drop a past run's smaller design into the slightly bigger slot?

While the quarter-slot projects would be compatible, the other slots wouldn't be. The bonding would be slightly different, and you need to make some changes to the layout to make it compatible with the larger slot size. Personally, I would keep the full slots the same size (@tnt from Tiny Tapeout will be happy), and adjust the half and quarter slots instead (rip off the band-aid once). As a nice touch, that way we would still be compatible with the OpenMPW shuttle projects. I'm not sure what Andrew meant by that. As I said, I just need to know the total saw street width.

Leo Moser (mole99)

As I said, we can go up to ~250um total saw street width. However, the larger the saw street, the less area there is for the projects.

Could we just make the 1x1 slot (and half slots?) grow slightly bigger to keep the quarter slot the same size?

In theory, yes, but this will increase the required area for the reticle. I'm also not sure if we can keep the saw street width at 200um without removing a column of projects then. If you want me to, I can explore that option.

That way you can always drop a past run's smaller design into the slightly bigger slot?

While the quarter-slot projects would be compatible, the other slots wouldn't be. The bonding would be slightly different, and you need to make some changes to the layout to make it compatible with the larger slot size. Personally, I would keep the full slots the same size (@tnt from Tiny Tapeout will be happy), and adjust the half and quarter slots instead (rip off the band-aid once). As a nice touch, that way we would still be compatible with the OpenMPW shuttle projects. I'm not sure what Andrew meant by that. As I said, I just need to know the total saw street width.

Total saw_street_width <= 100um I think we have that covered. Currently on the our reticle's edges we have saw_street_width/2 and when GF put their test features they did not honor any spacing, so the border around our reticle should be saw_street_width I hope I'm making sense here.

Andrew Wingate

Total saw_street_width <= 100um I think we have that covered. Currently on the our reticle's edges we have saw_street_width/2 and when GF put their test features they did not honor any spacing, so the border around our reticle should be saw_street_width I hope I'm making sense here.

Total saw_street_width <= 100um I think we have that covered.

Wasn't Tim just saying above that he would like to have 200um?

Currently on the our reticle's edges we have saw_street_width/2 and when GF put their test features they did not honor any spacing, so the border around our reticle should be saw_street_width

In both the G801 and G802 reticle layout, the border is already a full saw_street_width (60um).

Leo Moser (mole99)

Total saw_street_width <= 100um I think we have that covered.

Wasn't Tim just saying above that he would like to have 200um?

Currently on the our reticle's edges we have saw_street_width/2 and when GF put their test features they did not honor any spacing, so the border around our reticle should be saw_street_width

In both the G801 and G802 reticle layout, the border is already a full saw_street_width (60um).

Now that I am thinking about it, you are correct. There was another image like this one and the width there also looked more narrow than it was laid out. This is likely the same thing. I think we just make them whatever width the layout we're shooting for allows and I agree we shouldn't go so far that we lose any slots. Out of curiosity, what width does that put us at?

@Leo Moser (mole99) - I'm still a bit confused about why the slot sizes need to change? Is it just so that we keep the 1x1 slot being perfectly the same as 2*half+scribe?

Tim 'mithro' Ansell

@Leo Moser (mole99) - I'm still a bit confused about why the slot sizes need to change? Is it just so that we keep the 1x1 slot being perfectly the same as 2*half+scribe?

Yeah, I think the current sub_slots have the original saw street baked in sub_slot = (full - saw_street)/2

@Andrew Wingate - But that is a choice on our side -- not something that is a hard requirement?

Tim 'mithro' Ansell

@Andrew Wingate - But that is a choice on our side -- not something that is a hard requirement?

This is correct, but if we widen the sawstreets everywhere else this will have to follow suit. Currently assume we did not change all of it, we would all of a sudden have smaller streets on the subdivision streets.

@Andrew Wingate - Do you have a feeling / understanding of the "don't want to dice through test structures" complaint? Is it just because we have such thin scribe lines?

Tim 'mithro' Ansell

@Andrew Wingate - Do you have a feeling / understanding of the "don't want to dice through test structures" complaint? Is it just because we have such thin scribe lines?

The way it was phrased to me is that dicing through the metal loads up the wheel Loaded wheels create more heat and pressure. Causing some of the blowout backside chipping I was referring to before. The way it sounded (I couldn't ask every question and had to read between the lines) wheel dressing (scraping off the loaded bits and truing the wheel) happen at programmed intervals they can change. So dicing through them loads up the wheel more. That part is out of our control, and everything I've seen you produce since has had such wide margins in the areas they cared about most, I will say they won't mind dealing with it.

Leo Moser (mole99)

Nice! Yes, we'll have to switch to a different reticle layout, which reduces the overall capacity by a few quarter slots, but other than that, it should be fine. I'll update the reticle stitcher in the next days to give this layout a try, so we can be sure that it fits, and we know what the new sizes for the quarter- and half-slots are going to be.

Alright, I took the time to give this a try with the reticle stitcher. Here are the results:

7:57 a.m.

Reticle layout (90deg rotated)

Reticle sizes for different saw street widths (60/100/150/200um):

60um: 24.012 x 28.561 mm
100um: 24.292 x 28.821 mm
150um: 24.642 x 29.146 mm
200um: 24.992 x 29.471 mm

The maximum reticle size is 25 x 30 mm.

7:57 a.m.

Reduction in die area of the half- and quarter-slots

0p5x1:

60um: 1936 x 5122 um = 9916192 um² -> baseline
100um: 1916 x 5122 um = 9813752 um² -> -1.03 %
150um: 1891 x 5122 um = 9685702 um² -> -2.32 %
200um: 1866 x 5122 um = 9557652 um² -> -3.61 %

1x0p5:

60um: 3932 x 2531 um = 9951892 um² -> baseline
100um: 3932 x 2511 um = 9873252 um² -> -0.79 %
150um: 3932 x 2486 um = 9774952 um² -> -1.78 %
200um: 3932 x 2461 um = 9676652 um² -> -2.77 %

0p5x0p5:

60um: 1936 x 2531 um = 4900016 um² -> baseline
100um: 1916 x 2511 um = 4811076 um² -> -1.82 %
150um: 1891 x 2486 um = 4701026 um² -> -4.06 %
200um: 1866 x 2461 um = 4592226 um² -> -6.28 %

The quarter slot sees the largest reductions, due to the decrease in both width and height.

7:57 a.m.

@Tim 'mithro' Ansell If you'd like to go for a saw street width larger than 100um, I would suggest 150um. While 200um would still fit inside the maximum reticle size, there would be no space left to add anything else. Andrew mentioned that the dicing house wasn't too happy about the sliver of silicon with the test structures on it - we can't remove that, since the foundry adds it. However, we can add some dead silicon to all four borders of our reticle to make that sliver larger and thus less problematic. If we chose a 200um saw street width, there would be no area left to do that.

7:58 a.m.

The question now is whether we want to change the saw street width already for Run 3? I suppose there is still enough time until the tapeout in December. People will need to reimplement their designs anyway after the PDK update that I am going to release soon.

Leo Moser (mole99)

@Tim 'mithro' Ansell If you'd like to go for a saw street width larger than 100um, I would suggest 150um. While 200um would still fit inside the maximum reticle size, there would be no space left to add anything else. Andrew mentioned that the dicing house wasn't too happy about the sliver of silicon with the test structures on it - we can't remove that, since the foundry adds it. However, we can add some dead silicon to all four borders of our reticle to make that sliver larger and thus less problematic. If we chose a 200um saw street width, there would be no area left to do that.

To be clear there are 2 slivers. One has the test structures and that one is fine. The other was that very very small one. That was the one that was a real problem.

Andrew Wingate

To be clear there are 2 slivers. One has the test structures and that one is fine. The other was that very very small one. That was the one that was a real problem.

Is this Run 1? Both slivers have test structures on them. The very small one is also added by GF. The large sliver is 99% dead silicon - added by us.

8:10 a.m.

This is Run 1 with metal fill visible. You can see that sliver here at the top:

8:10 a.m.

In Run 2, we removed the dead silicon, so you should only see very small slivers there.

8:12 a.m.

Since we now know that small slivers are problematic for dicing, we could again add some dead silicon at the top, and also at the right side of our reticle layout.

Maybe dumb, but how wide are the test structures ? Is it a problem to leave them attached to the dies on the side ?

Leo Moser (mole99)

Since we now know that small slivers are problematic for dicing, we could again add some dead silicon at the top, and also at the right side of our reticle layout.

That would be good. I was asked if we had to have a sliver if we could keep it >= 1mm that should be fine. Let me try to take a pic, it'll take a moment.

tnt

Maybe dumb, but how wide are the test structures ? Is it a problem to leave them attached to the dies on the side ?

I'll try to get a pic of that too. Hang tight

Andrew Wingate

That would be good. I was asked if we had to have a sliver if we could keep it >= 1mm that should be fine. Let me try to take a pic, it'll take a moment.

We cannot keep it >= 1mm. That big sliver you showed above is only 600um in width. Edit: the 600um are our contribution. If you take a look at the big sliver, you will notice it has two colors: the golden part is from our reticle layout, the silver one is what GF added. If we choose a 150um scribe line width for our new reticle layout, then we have 358um remaining at the top and 854um on the right. (edited)

Leo Moser (mole99)

We cannot keep it >= 1mm. That big sliver you showed above is only 600um in width. Edit: the 600um are our contribution. If you take a look at the big sliver, you will notice it has two colors: the golden part is from our reticle layout, the silver one is what GF added. If we choose a 150um scribe line width for our new reticle layout, then we have 358um remaining at the top and 854um on the right. (edited)

There's nothing from GF side that would stop us from having no sliver, correct? Can we just split it between all the sawstreets and leave all die ~45um larger?

Andrew Wingate

There's nothing from GF side that would stop us from having no sliver, correct? Can we just split it between all the sawstreets and leave all die ~45um larger?

No, GF adds the slivers on all four sides. That's where the test structures are placed.

@tnt @Leo Moser (mole99) I have added a bunch of pics, tried to get the test features as well, towards the bottom. Leo you are correct there are test features on all sides..

Andrew Wingate

@tnt @Leo Moser (mole99) I have added a bunch of pics, tried to get the test features as well, towards the bottom. Leo you are correct there are test features on all sides..

FYI The test features are usually placed inside the scribe line and are destroyed during dicing. However, we can't do that because we create our own reticle and don't have access to the test features. This is why GF adds the test features around our reticle you can think of our reticle as one large die).

Leo Moser (mole99)

FYI The test features are usually placed inside the scribe line and are destroyed during dicing. However, we can't do that because we create our own reticle and don't have access to the test features. This is why GF adds the test features around our reticle you can think of our reticle as one large die).

I always understood that bit, but thought (hoped) it was localized to a single edge.

I see. Normally, you want to cover as much area as possible with test structures due to local mismatch of the wafer.

1

1. Saw street 60um is edge to edge of seal ring. Very little dicing margin for chipping with full thickness cut as I need less than 30um blade. Which is not recommended for full thickness cutting. Per discussion, please provide 80-100um saw street width if you want to run full thickness dicing.

10:48 p.m.

2. Chipping that touch the seal ring. Base on initial trial, I cannot check if there is damage under the seal ring.

10:48 p.m.

3. Small index island. If wafer is full thickness, the aspect ratio of the island will be high, this will cause it to shift and lead to chipping during dicing. Aspect ratio of unit width to thickness should not be smaller than 1:1; excluding saw street. Best if you can remove it totally.

10:49 p.m.

4. Some discoloration or corrosion on the pad.

10:49 p.m.

5. Backside chipping quality seems OK with slow feedrate. This is from the full thickness wafer.

10:50 p.m.

6. After we thin down the wafer, we see a lot of unusual backside chipping even at slow speed. Top side quality is okay. Currently I am not sure if this is wafer stress related or process related. To eliminate, this we can run DBG process.

10:51 p.m.

@here some of the pics from disco. They really spent quite a bit of time trying to work through these for whatever reason. See notes above.

What's "DBG" ?

tnt

What's "DBG" ?

Dice Before Grinding Basically you cut partially through the wafer, flip it over so it's face down on the tape and grind away the back until the dies are separated. They were trying to sell me on it, but we're looking to cut costs and are widening the kerf allowances so we can keep full thickness and they can use a larger wheel.

1

Andrew Wingate

2. Chipping that touch the seal ring. Base on initial trial, I cannot check if there is damage under the seal ring.

This is why some fabs place staircasing perimeter resistor just inside seal ring, connected to test pads. It is easy shorts/opens process monitor, and is a signal for crack detection post-singulation. (edited)

nmz787

This is why some fabs place staircasing perimeter resistor just inside seal ring, connected to test pads. It is easy shorts/opens process monitor, and is a signal for crack detection post-singulation. (edited)

Awesome! Thank you for the information/ insight. This would be interesting to see. Current assumptions are that we are going to increase the saw streets to just allow more area. Thanks!