Click here to go to our main page on packaging
Click here to go to our page on millimeerwave packaging (new for November 2020)
New for December 2020. Trade studies on how to package MMIC chips for millimeter-wave will continue into the coming decades. One thing for sure is that RF wirebonds need to be replaced with something with less inductance and radiation. Hot vias are one possibility. Bumped flip chip is another.
Hot vias are implemented in MMIC processes to route signals to the chip backside, as opposed to ground vias (which are "cold"). This topic seems to be trending as designers try to cope with wirebond inductance in millimeter-wave designs. The only tweak to the process is that the foundry has to allow you to pattern the backside metal to isolate the via from ground.
The inductance of a via hole is way less than that of a wirebond, see this page on wirebond inductance and attenuation. That is the principal attraction to hot vias.
What are the downsides to hot vias?
- You will need to pattern the attachment media (epoxy, or solder). A solder bubble in the wrong place will cost you an entire assembly. And this is likely NOT a reworkable process...
- The substrate you attach it to will set your maximum dissipation. Chances are that it won't have anywhere near as good thermal conductivity compared to a copper-moly tab. LTCC, HTCC, organic boards arenot great dissipators. Maybe a field of microvias in the ground plane can help.
This is not a new idea. It was patented back in 2001 by Lampen, Setzko, Kazior (a Darpa progam manager since July 2020), Kizner and Wendler at Raytheon: US Patent # 6,175,287 B1, "Direct backside interconnect for multiple chip assemblies". So show some respect to the inventors, uyou can use the acronym "DBIT" instead of "hot vias" when you are talking about thit interconnect scheme. Below you can see how a "DBIT" chip connects to a substrate. The "hot via" 110 sits on epoxy blob 52A to connect to trace 102. Gorund vias in the substrate (86) also acts as heat path conductors to the housing (not shown).
You have to admit, this is a cool idea. What prevented the patent assignee from deploying it? Here's some background, told by an Engineering Fellow that grew tired of company politics and bailed: what happens when you merge three large defense companies that all had GaAs fabs, and you take away two of them. Then you form a "one-company" working group to decide what the future of millimeterwave assemblies looks like. The one remaining fab comes up with a brilliant solution, but engineers from the other two groups have their own ideas and are still pissed about losing their sandboxes and decide they are not going to tool up for something "not invented here". Add in a powerless corporate engineering structure, and the result is a fair amount of missed opportunities.