zGlue anyone?

[_joost_]

Does anyone here have experience with zGlue (zGlue.com) technologies?  The site looks really slick but we all know how that works. I am interested to learn how simple/easy developing a chiplet based zGlue SoC was and if it is perceived to be cost effective. Do you see this sort of tech becoming available for small startups anytime in the near future. Trend or just hype? 

I realize the question is rather open - just looking for experiences and/or opinions.

[ataradov]

I don't get it. How is this different from the standard PCB with WLCSP ICs? This programmable fabric my be somewhat interesting, but I can't find any pictures of any designs that are not 3D renders.

They list a lot of applications, but it is not clear if this technology could be used in those applications, or is actually used now. I'd like to see real commercial products using this.

[_joost_]

Well, supposedly, zGlue's claim to fame is that you build your configuration for low/no NRE and get your very own custom chip. But yeah, thats why I am asking ;-)

[ataradov]

You can do this already with normal PCBs though. And the NRE is the cost of the PCB design. BGA interposer PCBs are nothing new and commonly used. So are PCBs with castellated vias. Potting that up into something that resembles  BGA/QFN package is a bit harder, but probably doable, especially is you are fine with just a blob covering the components.

The risk of this approach is that is it is not scalable. The company may go under and you are dean in the water. Or your product may be a hit and all of a sudden your are bottle-necked on them being able to make chips.

There is probably some niche use for this, but I don't see it going into all the places they list. It is just not practical.

[_joost_]

Good points. Anything at all in their advantage you think?  Speed, power budget?

[ataradov]

Speed is not going to be better than plain copper track. In fact, I think that interconnect would be limited. I would not pass USB HS or SDRAM signals though that.

They advertise the ability to completely disable certain parts, which may give an advantage in power, but most modern ICs that are expected to be low power, achieve quite low power numbers without disabling the power entirely.

It is possible that this technology is so revolutionary that I just fail to see good use cases. But they are failing to convince me with their site. And all the sample designs are behind login, and I could not be bothered to make an account even with a fake email.

Ok, I registered with a dummy email. The designer thingy is VERY slow in my browser. You have to pick the ICs from their library. Although the library is quite extensive, so that is probably not a problem.

From what I see you get LGA package for most (all?) designs. You can get the same LGA with regular PCB if you can forego dynamic trace configuration.

[_joost_]

and I could not be bothered to make an account even with a fake email.

   Totally with you on that!   I did see that “the amp hour” podcast has an interview with one of zGlue’s engineers - perhaps there is some insight there, will listen to it when walking the dog.   But yeah, everything behind a login account and no actual examples makes one wonder if this is  more than the famous does-it-all chip from the HTML series...

[ataradov]

The simplest thing they can do to showcase their service is to point to any commercial off the shelf hardware designed using it.

It is not even an NDA issue, anyone can go to the store and buy the product to tear it down.

[Doctorandus_P]

Speed is not going to be better than plain copper track. In fact, I think that interconnect would be limited. I would not pass USB HS or SDRAM signals though that.

Why do you make such premature assumptions if you don't even know what the technology is?
I would be cautious until I know more but they do not make that easy. Their website is too slick for me and I can't get a grip on it.

This video from the "Open Compute Project" channel on youtube may be similar technology. (I have not seen it)
OCPREG19 - Accelerating Innovation Through Chiplets and the Advanced Interface Bus (AIB)


... and there may be others.

[woofy]

Chris Gammell's interview with Dr Ming Zhang, CEO and founder.
https://theamphour.com/499-discussing-chiplets-with-ming-zhang/

[ataradov]

Why do you make such premature assumptions if you don't even know what the technology is?

Because I've seen how SDRAM routing can affect its performance. You are right, I don't know if their technology is sufficiently good to route high speed signals and have them be functional. But it will not be better than copper on a PCB.

I would be cautious until I know more but they do not make that easy. Their website is too slick for me and I can't get a grip on it.

And that's the sign of a startup designed to drain some VC money.

They may be on to something, of course, and this may be the future of technology. But for now they essentially have nothing tangible and useful in production. While micro-modules like this with WLCSP chips are very well tested technology.

[_joost_]

Well, I took the dog around the block while listening to "app hour podcast" on zGlue and have to admit it did not clear up how this tech is different from WLCSP tech. A startup cost of 25K was mentioned, zGlue supposedly has access to many chiplets (i.e. standard chips one can purchase off the shelf, variety of vendors) and there is no IP licensing to worry about.  I am not familiar with WLCSP technologies enough to understand and value that 25K startup at the right level or any of the other mentioned benefits. I do see that a few SoC vendors whom I checked with provide pretty much their own IP only for integration. So you'll have to shop within their store which may not give you exactly what you want. (not sure if this is true or common across the vendors).

It is interesting to understand these technologies better as they appear to become in reach of small(er) budgets. But for a small shop, it represents quite an investment of education time to figure out technology, cost, etc other benefits vs the issues and problems.

I did ran into this presentation on YouTube: showing zGlue's ChipBuilder. Looks slick and certainly easier than several other vendors' notes on how to build a SoC.

[ataradov]

This is a completely different thing. You still use WLCSP chips for this. What this technology does is make a dynamically reconfigurable equivalent of a PCB. You still mount the same standard WLCSP ICs (which you do have tho pick from their store, but they have a good selection, I assume they added most WLCSP chips they could find, since there is some really obscure stuff there).

I personally see no added value by the PCB being reconfigurable. One example they provide is ability to completely disconnect some ICs from the power lines decreasing current consumption. This is certainly a use case, but I'm not sure it is enough to even justify considering this technology.

There are probably some other benefits to dynamic reconfiguration, but they are failing at communicating them.

[_joost_]

So if I were to want/need to create a collection of chips on a substrate (using the term generically), in your opinion, zGlue would not add/provide an advantage over traditional SoC technologies?  Is the 25K startup cost somewhat equivalent from what I can find at a more traditional SoC integrator? 

[ataradov]

I see no real advantages. You still deal with their substrate that is then wire-bonded to the package with LGA pads (I guess other packages may be available too).

In case of a regular PCB, you just layout the PCB as usual and then do the same LGA footprint eliminating the whole wire-bonding step. And wire bonds would have to be covered in epoxy, of course, so one more step compared to  a PCB-based design.

If you don't include the design/layout labor costs (which would be the same in both cases), $25k is more than enough to have a few runs using traditional technology. You don't need any special integrators, you just layout the board as usual, you manufacture it as usual, you place components as usual. There are some details, like the substrate for that PCB will not be FR4 anymore. It may even need to be a ceramic of some sort. But generally it is not that different from your normal PCB assembly.

And all of this only applies if the space is at absolute premium and you must do a separate module. If you are just doing a commercial design (as they list in possible application on their site), then you just place those ICs directly on your main board and be done with it.

Also, if the substrate is actually made out of silicon, then how is it economical with this area? You are literally doubling the area of the silicon required for the project. And silicon cost is a significant part of the IC cost.

One of the feedback panel shows their contractor's (TSMC) feedback. How is this relevant to the technology? Ok, the company is nice to the contractors and orders a ton of silicon, good for TSMC, but says nothing about the technology.

And another panel is form DigiKey, who would not care what that is as long as it lets them sell more chips.

They are obviously trying to pad their resume.

This may provide an advantage to companies doing things like stacked die, where an MCU and a memory or a radio chip are placed in the same package. But again, traditional stacked die design is a well tested idea, so the uses for this technology are limited here.