editor's blog
Subscribe Now

Going Expensive to Reduce Interposer Cost

Imec has been  working 2,5D IC issues with a particular focus on optimizing costs and, in particular, test yields. Yields can take what might have been straightforward-looking cost numbers and make things not so clear.

In their work on interposers, Eric Beyne took a look at three different ways of routing the signals from a wide-I/O memory. These puppies have lots of connections – like, 1200 per chip. He explored three different ways of implementing the interposer to find out which had the best cost outlook. The idea was to connect two such interfaces, with four banks of 128 I/Os each. Each channel had 6 rows of 50 microbumps. Microbump pitch along a row was 40 µm; along a column it was 50 µm. The two simply needed to talk to each other on the interposer.

The cheapest, most traditional approach is to use PCB (or PWB) technology. An aggressive version would have 20-µm pitch and 15-µm vias. This approach resulted in an 8-layer board; you can see the layout below – lots of routing all over the place. Wire lengths were, on average, 180% of the die spacing.

 

Laminate.png

 

Next was a semi-additive copper process – more aggressive dimensions and more expensive. Line pitch was 10 µm; vias were 7 µm. the tighter routing allowed connectivity with only 4 layers, and the average wire length was 166% of the die spacing. You can see the slightly less colorful result below.

 

RDL.png

 

Finally, they took an expensive approach: damascene metal lines. Moving from the PCB fab to the silicon fab. But this got them down to 2-µm pitch with 1-µm vias, and that was enough to run wires straight across on 2 layers with no extra routing. In other words, wire lengths were equal to the die spacing. You can see this on the following picture.

 

Damascene.png

 

So what happens to the overall cost? The last one is nice, but expensive to build. And here is where yield comes in. Because the “most expensive” option uses only two layers, it has the best yield. And that yield more than compensates for the expensive processing, yielding the cheapest option.

 

They didn’t give out specific cost numbers (they typically reserve those for their participants), but the net result is that they believe the damascene approach to be the most effective.

 

 

 

Images courtesy Imec.

Leave a Reply

featured blogs
Jan 15, 2021
I recently saw (what appears at first glance to be) a simple puzzle involving triangles. But is finding the solution going to be trickier than I think?...
Jan 15, 2021
It's Martin Luther King Day on Monday. Cadence is off. Breakfast Bytes will not appear. And, as is traditional, I go completely off-topic the day before a break. In the past, a lot of novelty in... [[ Click on the title to access the full blog on the Cadence Community s...
Jan 14, 2021
Learn how electronic design automation (EDA) tools & silicon-proven IP enable today's most influential smart tech, including ADAS, 5G, IoT, and Cloud services. The post 5 Key Innovations that Are Making Everything Smarter appeared first on From Silicon To Software....
Jan 13, 2021
Testing is the final step of any manufacturing process, and arguably the most important, and yet it can often be overlooked.  Releasing a poorly tested product onto the market has destroyed more than one reputation for quality, and this is even more important in an age when ...

featured paper

Overcoming Signal Integrity Challenges of 112G Connections on PCB

Sponsored by Cadence Design Systems

One big challenge with 112G SerDes is handling signal integrity (SI) issues. By the time the signal winds its way from the transmitter on one chip to packages, across traces on PCBs, through connectors or cables, and arrives at the receiver, the signal is very distorted, making it a challenge to recover the clock and data-bits of the information being transferred. Learn how to handle SI issues and ensure that data is faithfully transmitted with a very low bit error rate (BER).

Click here to download the whitepaper

Featured Chalk Talk

Introducing Google Coral

Sponsored by Mouser Electronics and Google

AI inference at the edge is exploding right now. Numerous designs that can’t use cloud processing for AI tasks need high-performance, low-power AI acceleration right in their embedded designs. Wouldn’t it be cool if those designs could have their own little Google TPU? In this episode of Chalk Talk, Amelia Dalton chats with James McKurkin of Google about the Google Coral edge TPU.

More information about Coral System on Module