The waveguides, modulators, microscopic lasers, photodetectors, and other optical components that form photonic integrated circuits (PICs) all need to be perfectly aligned to minimize signal loss on the chip. PICs simply cannot improve yields, lower costs, or efficiently scale without ever-faster alignment tools able to perform with nanometer precision. In this episode, TECH B2B’s Dan McCarthy chats with Physik Instrumente’s Scott Jordan about the advanced alignment tools and AI algorithms that are helping to streamline the fabrication and packaging of PICs today.

Dan McCarthy: [00:00:06] Welcome again to the “Manufacturing Matters” podcast, where we talk about the trends and technology affecting global manufacturing. We’re recording on-site at Photonics West 2025. Thank you to SPIE for allowing us to be here. I’m Dan McCarthy. I’m with TECH B2B Marketing. Joining me today is Scott Jordan of Physik Instrumente (PI), who I have known for many, many years. Scott, first let’s kick off by, give me some background on Physique Instrumente and what your role is there.

Scott Jordan: [00:00:37] Well, Physik Instrumente is a German company. It is about 56 years old now, and it grew up with the semiconductor industry. It was founded by a couple of ceramicists who were skilled in this technology known as piezo nanopositioning. And so they could do submicron and sub-submicron and even nanoscale positioning back before there were really applications for it. And then things like the semiconductor industry and the need to manufacture billions of chips per year with increasingly small feature sizes and the whole Moore’s Law thing started to happen. And PI was in the right place at the right time. We’re headquartered in Germany, but we have a global footprint with applications, sales, and increasingly manufacturing, engineering, and innovation offices around the world. So, we try to be where the customers are to help them keep pace with the trends that they need to follow.

Dan McCarthy: [00:01:47] I see. So, a lot of your technology was in the steppers of the time for . . .

Scott Jordan: [00:01:52] Yeah. Motion control. Back in the early days, even things like encoders were sort of exotic at the time. Linear encoders were fabulously expensive. We made our own sensors for our micro/nanopositioners to allow us to control and stabilize motion down into the nanometer realm. And again, we were one of the first movers in this–in fact, maybe even the first mover in this field. Then the semiconductor industry started to reach a phase where manufacturing matters, and that meant more than just resolution. It meant things such as yield and throughput. These all became very important drivers for the industry, the whole notion of productivity and cost of ownership. These all evolved from those very early beginnings.

Dan McCarthy: [00:02:55] Okay. And before we jump into trends, your role is with photonics and nanopositioning. You tell us real quick how those two things converge.

Scott Jordan: [00:03:03] Well, you really need nanopositioning in order to do photonics efficiently. To connect two photonics devices is not as simple as just touching two wires together or soldering something together. No, they need to be precisely aligned down into, well, typical tolerances on the order of 20 nanometers or thereabouts. So, that’s where these two fields converge. Now there are many other applications of nanopositioning, and I’ve played in them in my time at PI in the life sciences, for example. It’s central to genomics. If you happen to have a tumor and the doctor wants to know what mutation you’re dealing with to choose the correct medication, that sort of thing, they can now do genomic sequencing affordably and quickly, in part because of PI’s contributions to that field, to making the the genomic scanning process possible in the clinical environment. So, nanopositioning means many things. But it is also absolutely essential for for photonics.

Dan McCarthy: [00:04:13] And manufacturing.

Scott Jordan: [00:04:14] Yeah, manufacturing and fabricating the processes. Because after all, today photonics is being driven by what’s called silicon photonics. This is a wafer-based technology. It is the semiconductor industry. So, once again, there’s this  convergence of capabilities that is happening as we speak.

Dan McCarthy: [00:04:33] Right. And so it’s important to align these things closely because unlike electricity, where it’s contact based, you have to make sure that the optical signal is jumping from one connection to the next.

Scott Jordan: [00:04:44] Right. Imagine you have two fire hoses, and you need to put them together to get the water from one end to the other. Alignment matters. So, think of an optical fiber or waveguide as being just a conduit for a fluid of photons. And in order to get those photons from device A to device B, you require this significantly sensitive multidegree of freedom and increasingly multichannel alignment and often on both sides of the device. Say you have a photonic integrated circuit. It has inputs. It has outputs. But when you move the fibers that feed the input, that moves the point of optimum alignment on the output, just like shining a flashlight through a, you know, the center of a roll of paper towels or something. You can you can steer that beam as it goes through it. So, that means alignment needs to be a cooperative process across degrees of freedom, inputs, outputs, channels.

Dan McCarthy: [00:05:48] And this was a big deal obviously years before where we’re talking about the telecom industry, we were connecting optical fibers to to devices. But now, as you’ve referenced a couple of times, you’re talking about PICs–photonic integrated circuits–or silicon photonics were installed onto the wafer base and you have waveguides. And when you package these things . . .

Scott Jordan: [00:06:06] Waveguides, modulators, little microscopic lasers, all those things, detectors, what have you. There are many fundamental devices that form these photonic circuits that are being fabricated alongside the microelectronic circuits. And they all have in common the need to be fed photons accurately with minimal loss. And that requires this alignment process. And it starts at the wafer level. Because most of the cost of building a photonic integrated circuit package is in the assembly in the packaging process. It’s about 80% of the cost according to various studies. And that means that you don’t want to package bad chips.

Dan McCarthy: [00:06:54] Good point.

Scott Jordan: [00:06:55] You’ll be wasting money. Burning it up. So, that means the chips must be tested, qualified, and increasingly “binned,” meaning graded for performance at the wafer level. So, you need to start aligning at the wafer level, and then the wafer is diced into its individual chips. It’s easy to kill the photonic chips, so it needs to be tested again, which requires more alignment. Then placing it into the package along with whatever else is being, you know, packaged with it over and over and over again, typically a dozen or two dozen times. The alignment process needs to be repeated. And of course, in the final assembly as well, you get the devices put together. So, alignment has been the big bottleneck. And that’s what our focus has been–to reduce the painfulness of active alignment, to make it something that just happens in a second or so. And that is what we’ve accomplished, is what we put into the marketplace.

Dan McCarthy: [00:07:58] Describe for me what the technology looks like. What do the machines you are are using look like? Is there different equipment for measuring at the wafer level versus the package level? What does Physik Instrumente provide in terms of systems?

Scott Jordan: [00:08:12] Well, we make subsystems. We make the . . .

Dan McCarthy: [00:08:16] Motion control.

Scott Jordan: [00:08:17] Yeah, if you go into the semiconductor industry and you look at tools that are totally unrelated to photonics, you have wafer handlings and you have positioning systems, you have metrology subsystems.

Dan McCarthy: [00:08:29] That hasn’t changed.

Scott Jordan: [00:08:30] Yeah. They put them together into a system, a tool. We make the microrobotic elements of a photonic tool, whether it’s for test or for assembly or something else. If it requires alignment in order to get the job done, we can do that super efficiently with modules that our OEMs can build into these, these tools. In a way, this photonics industry is exactly where the semiconductor industry was, say 1985 or so. One of my presentations, I have this this video from the Chip History [Center] museum that shows an Intel fab circa 1985. So, 40 years ago. And there’s all these people. And in back of them is custom equipment and manual processes and people talking and so forth. Just like photonics today. And all of that has been replaced. It’s been replaced because they want people out of the process. They need the throughput that they could not get from manual processes. And then, custom equipment is still important for the industry. But increasingly this ecosystem is rising. I mentioned wafer test, okay. Wafer probers were the first adoptions of this technology because imagine you have a 300 millimeter wafer. It’s got thousands of chips on it. They all need to be tested. If it takes you 5 minutes to align to each one, forget it. Production economics has gone out the window. So, our technology brings that down into the second realm or even subsecond round, which makes the whole economics work.

Dan McCarthy: [00:10:15] Okay. I know that part of the economics in this general trend is talk about passive alignment as well as active alignment. You’re on the active side. Passive is where the alignment is sort of baked in to the wafer. There’s some problems with that too.

Scott Jordan: [00:10:30] Well, assive alignment has been the unicorn of this industry for as long as I’ve been involved, and that’s more years than I really care to think about. But it’s it’s . . . people have been chasing it. A few people have come up with really good ideas, like there’s a group called Teramount out of Israel that has a very clever approach to connecting things together in photonic scale, but these are specific to certain applications. We also have other customers, other contributors to the field, who have developed passive alignment technologies. But there’s nothing that would solve the problem, for example, of wafer probing. Wafer probing. I have a wafer. I’ve got thousands of devices on it. I can’t be saying, well, each of those is like a little Lego kit, and I can just plug the Legos in. You can’t do that at wafer scale. You need to do things in a noncontact manner and that requires active alignment. After all, what has been driving passive alignment is how painful and costly active alignment has been. So, if you take the painfulness and the cost out, that reduces the the need, the urgency to have a passive alignment process. I think there’s room in the world for both.

Dan McCarthy: [00:11:54] I would agree.

Scott Jordan: [00:11:55] Both technologies. And as this industry expands, passive alignment technologies will and are arriving to solve certain specific problems, but they can’t solve them all. And in the meanwhile, if we have taken, if we, PI, have taken the pain and the cost out of active alignment, doesn’t that make life easier for everybody?

Dan McCarthy: [00:12:22] Yes.

Scott Jordan: [00:12:22] I think so.

Dan McCarthy: [00:12:23] So, you talked a little bit about, active alignment. I’ve lost my question. Active alignment requires motion control. Is the solution largely in the mechanics? Is it software or is it computational? As you get tighter and tighter tolerances, where do you have to make those innovations?

Scott Jordan: [00:12:45] Well, here I’d like to make a general comment about the motion control industry. I mentioned earlier that in the early days when PI was just getting started, things like encoders were precious commodities, and high performance controllers didn’t really exist, and those all evolved over the years. And at the same time, costs came way down. It used to be a linear encoder was a really elite element, and it made for a very expensive translation or rotation stage. It was a rotary, obviously, a rotary equivalent of a linear encoder in that case. Those have all come down dramatically in cost over the decades. So, it’s no longer an issue. In fact, it’s almost a commodity situation to get adequate stages. Okay. What is the future of motion control? I think it’s the same as for the future of automobiles. It’s autonomy. It’s intelligence. It’s the algorithms that are built in that are there to accomplish a task. So, for example, I can get in my car, and if it’s a Model T, I have to shift. I have to turn the steering wheel. I have to brake. I have to pay attention. Okay? I have to do it all. But if I get into my Tesla, I can say, take me to Aldi and off I go. So it accomplishes a task. Now imagine the task is align my device. Issue a command and it happens because of the algorithms built in. Yes, the mechanics are important. Yes, the electronics are important, but the magic happens at the algorithm level. And that’s where we have been driving that, that trend at PI.

Dan McCarthy: [00:14:39] So, were these algorithms. Would this be deep learning, machine learning, AI type algorithms or are they . . .

Scott Jordan: [00:14:45] We are leveraging AI for our latest algorithm, which is a stunningly fast capability for achieving first light, getting those first photons through. Optimization–we nailed that long ago. In 2016, we won the R&D 100 award for solving that problem.

Dan McCarthy: [00:15:09] Right.

Scott Jordan: [00:15:09] Multivariate, multielement, multichannel alignment, making it all fast in a single command. Well, now there’s still the problem of what if I have no light at all? What if I’m millimeters away from alignment? How do I get that first glimmer of light through? That’s the problem we have recently solved, and that involves . . . I can’t say too much because of the patent process and all that, but it involves the latest in compute capabilities and tools and functionality. It’s so, so cool. It’s just the beginning.

Dan McCarthy: [00:15:48] Okay. Yeah. You can’t really start calibrating the strength of the signal, right, until you have that first light. So that is important.

Scott Jordan: [00:15:54] You can tell I love talking about this sort of thing because it is so cool, and it’s an enabler for the whole industry.

Dan McCarthy: [00:16:00] It is, it is. That’s one of the things we talked about very early in our relationship that that alignment is, in packaging, is really where PICs will make or break it. I mean, they’re developing higher yields. That’s going to be a process that is somewhat familiar from the semiconductor industry. But packaging optical components and making those connections is really what distinguishes it. And that’s where a lot of the innovation has to happen. So, it has to happen faster. Otherwise this is just not going to . . . you’re talking about, the scales that they need to do to make this all economical. Speed is key. So, on that note, these components are becoming increasingly dense. They’re putting more photonic components on the wafer, on the chip, packaging, there’s more. I mean, you’re limited by real estate somewhat, but there’s more connections to the package as well. As that happens, you’re going to have to keep pace with that. What are some of the innovations you’re working on now that’s going to help keep pace with that?

Scott Jordan: [00:16:54] It’s all driven by the customers. You mentioned the number of connections going to the packages. We’ve seen some amazing stuff coming out of companies like Avicenna and RAM Photonics, where they have these arrays with hundreds or even thousands of optical fibers. Aligning that is a challenge. And so, we make it easy. Really. It’s no more difficult for our equipment to align something with a thousand channels than it is to do for just, you know, two channels or four channels or something like that. It’s all the same command that you issue to our controllers, and it happens. One of the trends that we’re seeing is, as this industry becomes more production focused because manufacturing suddenly matters to photonics, then we see an ongoing and increasing need for different form factors, different travel ranges, that sort of thing. The applications are becoming more diverse and so must be the, you know, the implementations where what you saw us come out with in, I mentioned 2016, that was largely the big winner that in that era was this hexapod assembly with a piezo fast piezo nanopositioner on the end.

Scott Jordan: [00:18:31] So, you have the six degrees of freedom. You have all the fast algorithms that get the job done quickly. That became the industry standard very quickly, but now we’re having customers come to us and say, I want to commission a custom tool where I need something this big and it has to fit inside, has to be embedded, and it needs so however many degrees of freedom, and the algorithms and all that stuff. So, at this conference, we’re actually showing our answer to those challenges. But our whole product line is involved. We have customers who need large gantry systems to accommodate large substrates or carriers, trays. And now we can do that. We have moved well beyond the hexapod with a nanopositioning device on the end to the same algorithms. The same capabilities can be mixed and matched to address specific custom needs that customers have. And it’s working great. It’s, it’s solving problems for our customers in a very rapid manner. Being able to do things so, so in such a modular fashion is . . .

Dan McCarthy: [00:19:47] And to your earlier point, as chips become more specialized, you are going to need something as versatile as active alignment tools. You can’t bake solutions into every chip because every chip is going to be application specific.

Scott Jordan: [00:19:58] So, well, exactly. Everything has to be versatility, modularity, even such mundane things as as serviceability. Things like that. Just being able.

Dan McCarthy: [00:20:11] Mundane.

Scott Jordan: [00:20:12] Yeah. Well, also we have one customer, who’s a very big company, say that they like our, you know, the stacked approach because that way if they have a need where they suddenly next week they’re going to need twice as much traffic in one axis. So, they can just pull the thing apart and put a longer stage on that one axis and, and proceed from there. So, modularity has a lot to do with it. It’s hard to predict the applications in a field that is moving so rapidly. And the ability to respond quickly with this adaptable architecture, I think is key.

Dan McCarthy: [00:20:53] I don’t want to leave the podcast leaving unasked questions on the table. So, is there anything I didn’t cover on this? Any future trends that you see in PICs, particularly? I know data centers are the big market, for example, we didn’t talk about different types of PICs, lidars. I’m sure the challenge is still the same. It’s still always going to be optimizing the signal more quickly.

Scott Jordan: [00:21:17] I’d say that the ongoing challenges for the photonics industry in the semiconductor industry, which are being mashed together, there’s lots of lessons to be learned on both sides. Right now, if I walk into a silicon photonics operation, there’s going to be a mixture of photonics people and semiconductor people in the room. It’s easy to tell who is who.

Dan McCarthy: [00:21:39] Because they’re standing apart.

Scott Jordan: [00:21:40] Yeah, because the semiconductor guys, they are battle hardened. They know all about how to, you know, deal with production economics. And, and now they have this, this amazing new technology that is something that can be put on the same wafers alongside the electronics or instead of electronics. You know what it’s all about the interconnects. Everything that we’ve been talking about is about connectivity for people. We’re all walking around with supercomputers in our pockets nowadays. AI has become a thing. As we record this this week, the big shift, big controversy has been the new DeepSeek algorithm out of China. It’s open source.

Dan McCarthy: [00:22:26] Less complicated chips.

Scott Jordan: [00:22:28] Yeah, yeah. Well, it’s interesting to compare to, to take this news about DeepSeek and compare it with Apple’s strategy that they announced or described not that long ago, just a few months ago. They are committed to doing things on the device. Well, how impractical is that? You know, considering it takes these huge banks of high-end chips to do the AI trick. Well, DeepSeek is the first glimpse at a future where this is possible on practical consumer devices. I think it’s great news for the industry. People are saying, oh, well, now, you know, it’s bad news for NVIDIA because they’ll sell fewer high-end chips. I don’t think that’s necessarily the case because, when was the last time you saw a commodity go down in cost and didn’t exponentiate in demand?

Dan McCarthy: [00:23:21] That’s an excellent point.

Scott Jordan: [00:23:22] Well, that’s what we’re looking at. And frankly, it involves massive amounts of data to be . . . You need a large language model on your smartphone. That’s gigabytes, tens of gigabytes that needs to be sent to your phone and refreshed and very often with feedback back to the mothership. Immense amounts of data. And that’s good news for the photonics industry. That’s why the data center market is not going to slow down anytime soon.

Dan McCarthy: [00:23:54] I would agree.

Scott Jordan: [00:23:55] And then, autonomous cars, which is a subclass of artificial intelligence, where you get into your car and say, take, take me to Aldi and it drives you there. Today, 10 exabytes of data per month is sent over the internet regarding autonomous vehicles sending information back up to the mothership to train the models, sending the updates to the cars themselves. Ten exabytes. Probably mostly Tesla at this moment. It’s only going to grow. So, right there you have an example of where AI is becoming embedded in a consumer device, which is the automobile. And the consequences of that are really good for the photonics industry. And we’re only going to see more of that as the costs come down.

Dan McCarthy: [00:24:50] That’s right. That’s a good point. Yeah, I think that’s been one of the sticking points is demand hasn’t quite been there to justify the investment in scaling up. But I think that’s changing on several fronts. People always talk about the killer app. I think there are actually several coming up.

Scott Jordan: [00:25:05] Never, never bet against the democratization of technology. It’s just the natural flow of things. Water runs downhill in the same way technology gets cheaper and more broadly applied.

Dan McCarthy: [00:25:18] All right. Well, Scott, appreciate your time on the show today.

Scott Jordan: [00:25:22] Well, thank you.

Dan McCarthy: [00:25:23] And wish you luck. I look forward to seeing what PI comes up with next year when we come back.

Scott Jordan: [00:25:28] Just watch. We got some cool stuff coming.

Dan McCarthy: [00:25:30] That’s awesome. Well, thanks everyone again for joining the “Manufacturing Matters” podcast. You can find us on all your favorite streaming podcast channels. You can also find us at manufacturing-matters.com. If you have any questions for Scott, please feel free to send them to us. We’ll forward them on. We’re recording here live. Not live. Sorry, but we’re recording it on-site at Photonics West 2025 and hope to see you again soon. Thanks.

Scott Jordan: [00:25:55] Thank you.