Unknowns And Challenges In Advanced Packaging

Dick Otte, CEO of Promex Industries, sat down with Semiconductor Engineering to talk about unknowns in material properties, the impact on bonding, and why environmental factors are so important in complex heterogeneous packages. What follows are excerpts of that conversation.

SE: Companies have been designing heterogeneous chips to take advantage of specific applications or use cases, but they also need to design advanced packages for those heterogeneous components. Do tools exist to make this work smoothly, because a lot of these are custom designs?

Otte: There are a lot of powerful tools out there. If you look at design software, for example, SolidWorks does a great job for the mechanics. It will go down to sub-micron levels and handles it just fine. And there are expensive software packages for finite element analysis in great depth, and the computing power we have developed makes it viable to do that and get results quickly — sometimes in minutes. But we do not know enough detail about the properties of materials to use them effectively in designs. That’s one of the weaknesses. If you look at a standard data sheet, you would be lucky if it tells you the modulus of elasticity, let alone what the glass transition temperature is and how the modulus changes as you roll through it — especially when combined with the coefficient of thermal expansion. You need to know all that stuff to truly “design” it. The next thing you get into is how do we get the parts? 3D printing is coming along and is helping a lot to improve the availability of parts. But it has some limits, especially surface finish. Today, most of the processes for doing 3D printing give you a relatively coarse finish, somewhere on the order of 300 micro-inches that you can make with that RMS [root mean square] measure of surface finish. That’s too coarse, especially for optical. It’s great for adhesion, because things really stick to these little hills and valleys, but there’s a need for very flat surfaces.

SE: This is basically the equivalent of line-edge roughness at very advanced nodes, right?

Otte: Yes, and that’s a good analogy. There also is a lot of effort being spent around MEMS. While MEMS traditionally has been done in silicon, we’re starting to see those processes utilize other materials. Etching of metals has been commonly done on sheets. And so where can we go with those kinds of lithographic techniques to get really high-resolution parts? And once you get the parts, how do you join them together. You may have a glass part, a plastic part, and a metal part. How do I join them? The first problem you have is getting sufficient adhesion to the surface. Then, once that problem is addressed, how do I keep it together when I thermally cycle it? If the ceramic has got a 3 ppm coefficient of expansion, and this polymer I use because I like its optical properties is at 50 ppm, what happens when I cycle it to 150°C to meet MIL-STD-883? The answer is that it comes apart.

SE: What other challenges are you seeing?

Otte: Once you get the device assembled, how do you get it to tolerate the environmental requirements. And what are they? There’s a tendency for people to go back to the old Bell standards, MIL-STD-883. The cell phone guys gave up on that a long time ago. They’ve got their own criteria that’s relevant. One of the key things designers of commercial devices have to contend with is whether the requirements are real for their environment? In Florida, there is high humidity. In Saudi Arabia, you have wind and dust and high temperatures. And these devices need to work in the Arctic, too.

SE: So the ambient temperatures are critical?

Otte: Yes, but so are humidity and water absorption, which often are not well addressed in material data sheets. What is the impact on dielectric constants? What are the physical dimensions and is there a tendency for it to blow up when you heat it? After all that, when you get the product out into the real world, you still can run into unanticipated consequences, like the kid who swallows little batteries. How do you plan for that?

SE: And when you package different components together, there are a lot of variables. How do you address that?

Otte: You have to start with the basics. What are the characteristics of these parts? What am I trying to accomplish in terms of connections? There’s almost always a workable solution, but you pay a price here. For example, wire-bonding is still one of the most widely used interconnect methods because it’s so versatile. You can run a little wire from one place to another and it works. The bad news is that you have parasitics, the devices are fragile, and you have to fit it all into a limited space.

SE: Is there sufficient training and talent to deal with that?

Otte: No, and one of the issues we see is the people coming out of the universities today are incredibly talented in dealing with software and apps. What they don’t have are the experiences my generation had in using their hands to build things. I attribute much of my engineering skill to my hobby of building model airplanes when I was in grammar school and high school. I learned all about that, and I learned that if I wanted to fly my airplanes on Sunday, I had to get my act together early in the week and build it. So you learned how to manage projects and you glued all this stuff together. Today, there’s not nearly as much of these hand-crafted hobbies, and one of the limitations of that is that the younger engineering crew doesn’t intuitively understand the physical world. They’re more dependent upon the computer and analysis and design. It’s another way you can go, and it works well if you have the imagination and tenacity to fight your way through it, but it’s a different route to success.

SE: Because they’re working at higher levels of abstraction?

Otte: Yes, and it’s because in the past all you had to do was make the lithography finer and finer. We’ve really pushed that to the limit. One of the things we’re see here at Promex, because we’re doing heterogeneous integration and assembly — we don’t fab wafers — is an incredible amount of innovation in areas like medicine and biotech. People are making devices and developing solutions that utilize electronics to gather and process information and report the results. Their devices have to interact with people in the real world, analyzing blood and saliva, so they need sensors that incorporate the non-electronic parts. They’re also doing things like DNA sequencing, where you need to apply chemistry.

SE: That hasn’t worked out so well for some companies.

Otte: You need to focus on the physical details of what’s happening. It’s not just about finances and software. The real question you need to ask is, ‘Does this thing work?’

SE: Isn’t that one of the big challenges with chiplets? It’s not just soft IP anymore. You need to prove it in silicon.

Otte: That’s correct, and it takes a lot of skill and a lot of capabilities. You need a scanning electron microscope, and micro-probing, and a lot of analytic techniques that are not trivial.

SE: And AFM and multi-beam inspection?

Otte: Yes, and that’s why we’ve made a lot of investments in metrology. We have fully automatic optical comparators that are all electronic, which will make measurements down to the micron regime. And we have two Keyence devices. We use one for measuring flatness, because when you take 256-pin BGAs and you want to put them on a substrate, they’d better be flat to one part in 10,000 or they’re not going to join together. This tool will measure to that level of accuracy and tell you if things are going to work or not.

SE: How do you deal with warpage?

Otte: One of the things people don’t want to talk about in silicon is that now we routinely thin wafers to 100 microns. That’s very common. We do that every day, and you can go down as thin as 10 microns, and all of the magic of semiconductors still works because they’re in the top surfaces. But we don’t have the methods to handle the wafers and die once you get much below 50 microns. You can do 50 microns if the die are small. You can handle cm² die. But when you get that thin, it really gets tough.

SE: This should make for some very interesting engineering challenges over the next couple decades.

Otte: Yes, and there’s no end in sight to where it can go.

SE: How are we going to ensure these devices will work throughout their expected lifetimes?

Otte: You have to stay in touch with the user. They’ve got to tell you what is happening. You can make all kinds of predictions, but what counts is what you learn with time and experience of working with your customers. If you want to tell your customer that a device will last 10 years, you need to be able to show you’ve been building things using the same methods, with this surface-mount solder and these pad dimensions and these kinds of encapsulates, for 20 years — and that they work. When you start driving things like chiplets, what are the unintended consequences of chiplets? I can tell you all kinds of things that will cause a device to fail. The trick is to find a combination of things that work in the long run.

SE: One of the big challenges here is that you need the whole supply chain to be on board with this, rather than just individual pieces, right?

Otte: Yes, and we’ve been talking about what are the implications for the end product. There’s a whole other story underneath that about what’s happening with the equipment and the processes. What kind of tools and equipment do we need that we don’t have today. There’s a whole new emerging world. Where is that stuff going to come from, and who’s going to engineer it? Everyone wants to design the sexy next phone, but who wants to design the machine that’s going to put all this stuff together? We need equipment to do fundamental stuff, like put down a thin layer of silver on a substrate to get uniformity with no density or pinholes. And what happens when you try to put down alloys? This is really complicated, and you need a very good understanding of physics.