Advantest Talks Semi

Are Silicon Photonics the Only Way Forward in Semiconductors?

• Don Ong, Head of Innovation for the Advantest Field Service Business Group • Season 2 • Episode 18

Share episode

Silicon photonics is not just a buzzword—it's the future of the semiconductor industry, and we're here to uncover its secrets with Dr. Lee Chee Wai.

Dr. Lee Chee Wei is a distinguished expert in silicon photonics integration, boasting an extensive career spanning over 18 years. Holding a PhD in Photonics Integration from Nanyang Technological University (NTU) Singapore and a postdoctoral fellowship from the Cavendish Laboratory at the University of Cambridge, Dr. Lee has honed his skills both in academia and the industry. His academic journey was supported by prestigious awards, including the ASEAN Scholarship, ASTAR Graduate Scholarship, and ASTAR Overseas Postdoctoral Fellowship.

Currently, Dr. Lee is pivotal in advancing the commercialization of data center silicon photonics. His technical acumen extends to the design of photonic integrating circuits, and the development and design of advanced fabrication and packaging technologies. Dr. Lee's contributions to the field are highlighted by his extensive portfolio of more than 30 internationally filed patents and over 50 peer-reviewed journal publications. His work bridges the gap between east and west, showcasing a global impact in Photonic Integrated Circuit (PIC) Technology.

Discover how the convergence of light and silicon is about to shatter the limitations of traditional CMOS technology, unlocking unparalleled data speeds and efficiency, crucial for the demands of Gen AI and modern data centers. Dr. Lee guides us through the transformative potential of merging photonic and electronic components on a single chip with EPIC Electronics.

Our conversation takes a serious look at the nuts and bolts of integrating silicon photonics with CMOS. We confront the challenges head-on—material incompatibilities, efficient signal conversion, and heat dissipation—and discuss cutting-edge solutions like co-packaged optics and heterogeneous integration that are setting the stage for future advancements. With a focus on scaling these innovations for commercial viability since 2017, we discuss how these developments could redefine the semiconductor landscape. Tune in to grasp how silicon photonics stands as a beacon of promise, ready to reshape data centers and elevate the semiconductor industry with unprecedented speed and efficiency.

Thanks for tuning in to "Advantest Talks Semi"!

If you enjoyed this episode, we'd love to hear from you! Please take a moment to leave a rating on Apple Podcast. Your feedback helps us improve and reach new listeners.

Don't forget to subscribe and share with your friends. We appreciate your support!

Don: 0:01
Hello and welcome to another exciting episode of Advantest Talks Semi. I'm your host, Don Ong. In today's episode, we are diving into the fascinating world of silicon photonics and EPIC Electronics: Photonic Integrated Circuits. As the semiconductor industry shrinks down to two-nanometer scales and beyond, we're reaching a point where traditional CMOS technology is struggling to catch up with the demand for speed, efficiency, and data processing power. That's where silicon photonics comes in. By using light to move data, it's opening up new possibilities for faster, more energy-efficient technology. And then we have EPIC, where it takes things further by combining both photonic and electronic components on a single chip, allowing for even more powerful integrated solutions.

Today I'm thrilled to welcome a very special guest, a fellow alum from both NTU, Nanyang Technological University, and the University of Cambridge in the UK, Dr. Lee Chee Wai. Chee Wai is the CTO of a stealth-mode startup that's making impressive strides in silicon photonics, specifically around the next generation of EPIC. His journey through academia and industry is fascinating. After completing his PhD at NTU and a post-doctorate fellowship at Cambridge, he spent eight years as a research scientist at ASTAR, working on cutting-edge science in Singapore's top research institution. He then transitioned into engineering leadership roles at several companies before diving into the startup world. Chee Wai, it's fantastic to have you with us. Thank you for joining us.

Chee Wei: 1:49
Thank you, Don. Hi, everyone, I'm happy to be here.

Don: 1:51
All right, Chee Wai, to start things off, can you give us a simple overview of what silicon photonics is and why it's getting so much attention in the semiconductor industry?

Chee Wei: 1:55
I'll keep it simple. Traditionally, we have been talking about electronic ICs, which transmit electrons as signals. But for silicon photonics, it's actually using the same CMOS technology to fabricate what we call the silicon waveguide on the same wafer. The wafer itself can transmit light signals by routing and manipulating the light on the silicon chip. So that is what we call silicon photonics, and why it has such a huge impact on society is because, by using light, we can increase data transmission capacity by many, many times. That's something that electronic counterparts cannot do.

Don: 2:42
And then we talk about EPIC, electronic photonic integrated circuits, along with silicon photonics. So can you explain what is the connection between the two and what makes EPIC different?

Chee Wei: 2:54
EPIC itself means electronic photonic integrated circuits, and silicon photonics is just on the photonic integrated side. But it actually cannot exist alone. It has to combine with electronics. Because if you look at it nowadays, the application for silicon photonics is actually a combination of photonic IC and electronic IC. Even though photonic IC can promise a very high transmission speed and also a huge data capacity, it has to also work together with electronic IC to achieve the same function.

Don: 3:31
So silicon photonics is very often described as a breakthrough, like you mentioned, for data speed and energy efficiency. So can you help us understand why it is so important, especially when we push for semiconductors to go to very, very small scales?

Chee Wei: 3:51
When we are going from CMOS down to one nanometer, it's already reaching a bottleneck. So if you want to increase the speed further, you can't actually go into the same path for electronics. That's why people are switching into photonics. But photonics itself is a bit different from electronics in the way that it does not require very advanced technology nodes. So it only requires like 45, 60, 90, or 130-nanometer technology nodes. It’s something that you can actually fabricate in a very conventional foundry at a lower cost. But it actually can achieve huge benefits that I mentioned just now compared to electronic ICs.

Don: 4:38
Yeah, I see. So for somebody that is very new to this, right, for silicon photonics and EPIC. So what are the main challenges that silicon photonics and EPIC are designed to tackle? Maybe you mentioned data centers and advanced computing. Can you tell us a little bit more about that?

Chee Wei: 4:55
When you talk about electronic ICs, the bandwidth is already reaching the bottleneck. For photonic ICs, you can transmit light in, for example, one single fiber up to 80 channels at the same time. That's how it increases data bandwidth, speed, and capacity. So if you talk about data centers, they require a lot of data to be transmitted together at very high speeds. Nowadays, people are working on 400G, and in the near future, they’re moving to 800G and possibly 1.6T. That’s why you cannot rely on just electronic ICs anymore. You have no choice but to switch to silicon photonics. It’s the only way forward.

Don: 5:59
Diving down that area right. So it's very interesting because, as we have Gen AI, data centers need faster speed right now between our processor and memory. So why do you think this is the right time for Silicon Photonics and for EPIC to really take off? Because we know photonics is not new, it's been around for a while. But why right now?

Chee Wei: 6:19
You will talk about the history of photonics. I see it actually dated back to the 1950s, where it actually happened. The research started together with electronic IC. But just that, the electronic IC developed much faster and nowadays is very mature. But for following, IC is gaining less attention. But, like what you said, the electronic part has already reached a bottleneck. So people are looking for a new way and, yeah, here comes silicon photonics. But it was like when I started my research on silicon photonics it was relatively immature, only until like 2017, when the first silicon photonics product was commercialized by Intel. That's where people realize that actually silicon photonics can be put into an actual product and nowadays is gaining a lot of attention, no matter like any corner of the world, like from the east or from the west, people are diving into these silicon photonics.

Don: 7:28
So while mainstream semiconductors still rely on CMOS technology, silicon photonics is expected to enhance a few critical components, where number one reducing power consumption and increasing data transmission speed is very key. So it's essential to see how silicon photonics can work alongside CMOS devices. So can you explain why integrating silicon photonics with CMOS technology is so important, and what does combining these two really make possible?

Chee Wei: 7:58
If you talk on the technology side, the reason why people say that the silicon photonics is CMOS compatible is because silicon photonics can be fabricated in the same foundry without the need for a new kind of fabrication technology and also it requires a lower technology node. So that's the advantage for silicon photonics. But if you talk about combining them together, it's not that straightforward. Because when you talk about CMOS IC, the substrate that people are using—I'll give you an example—the insulator layer. You will be required to have a thinner insulator layer. But for silicon photonics, you need a very thick insulator layer so that the light doesn't leak into the substrate. So it actually presents two very contradicting requirements and because of this, it's very difficult for you to actually fabricate them on the same die.

Chee Wei: 9:00
And there comes the advanced packaging. So a lot of companies are pursuing something like the cold package optics, which actually through advanced packaging and combining or package the silicon photonics die together with the CMOS IC die, such as the ASIC or CPU, together on the same package. So this is something that people are looking into. Actually, a lot of companies are working on this. They call it the CPO. Yeah, that's the way to go.

Don: 9:33
So you mentioned one of the challenges, right. So, having silicon, photonics, using silicon, with CMOS, we're able to, using the same foundry, put it on the same silicon, and now the challenge is packaging, trying to put it together. So what are some of the other technical challenges that we're seeing when you try to bring photonics and CMOS together?

Chee Wei: 9:55
Okay, of course you talk about electronics is working on electronics, right, and the photonic IC is working on light. So how do you actually convert them together efficiently without loss? That is something that is a huge challenge for silicon photonics. It's actually one of the biggest challenges for silicon photonics is that the loss on chip is very high. That means the loss for the light signal is very high. So this is something that a lot of foundries are trying to tackle, such as to improve the process so that the etching, the plasma etching, can be done smoother and reduce the propagation loss for the light. So this is something one of the main challenges.

Of course, another challenge is how do you generate a signal on the silicon photonics IC, because silicon itself is not a direct bandgap material—this is something physics, yeah—so it cannot generate light. So you rely on other materials to generate the light signal, such as the 3-5 compound semiconductor. And how you actually integrate this kind of material onto silicon photonics is another challenge. And of course, now some companies are working on solutions for this kind of technology called heterogeneous integration, such as Intel. They are wafer bonding, wafer bond this kind of material—3-5 material—onto silicon photonics. And of course, there are other methods to do so, but this is one of the solutions, but we still have to look for more solutions like this to integrate different types of materials, such as 3,5 and also the tinfoil lithium niobate, and these are one of the other challenges that we have to overcome.

Don: 11:59
Great. So we'll dive a little bit more into the components later, but I'd like to hear about your opinion on reducing power use and boost data speed. So we often hear that this integration is going to help us get lower power consumption and faster data speed. So we often hear that this integration is going to help us get lower power consumption and faster data transmission. Can you help us break it down into how it actually works?

Chee Wei: 12:23
So, first of all, I'll give you an example, like for a data center transceiver, the optical components actually only take up 30% of the total power consumed and the electronics part, such as the DSP or the driver, actually take up the rest 70%.

So this is how the power consumption ratio between the optical and electronic components for just one data center transceiver. So how you actually realize this is because for the photonics part to actually modulate the signal to very high speed it actually can be done very quickly and it doesn't require a lot of channels. For example, to realize a 100G modulation pump 4 modulation you only need one channel on the silicon photonics die. But to realize this in a driver or DSP you only need one channel on the silicon photonics die. But to realize this in a driver or DSP you require a lot of channels combined together. Doing this kind of time division, multiplexing, yes, that's why it actually consumes less power. And for silicon photonics nowadays people are already working on 200G pump for modulation per channel. So if you combine, for example, you combine four channels, then you become 800G. So this is how simple it is for the photonics side.

Don: 13:52
Yeah, so I think, like you mentioned, in silicon photonics you're using lights, which has different wavelengths and they are by nature already segregated, yeah Right, so you get different channels. So on electronic signals, we are really on the same frequency, so that means we need to have physical different channels to help drive this data speed. So that's good to know. From a manufacturing perspective, how practical is it to scale things up for this integration, especially for, like data centers? So, using silicon photonics, how do we drive things faster? Is it practical, is it economical?

Chee Wei: 14:31
For transceivers. Nowadays they are already implementing the photonic chip onto the transceiver. So that's how they actually can keep on increasing the speed 100G, 200g, 400G and so on. So how they do it is actually they have a lot of this kind of pluggable transceiver on the server. Whenever they wanted to change the speed, they can actually just plug and play a new transceiver. So that's how you actually can scale up relatively easier, instead of replacing everything together.

Don: 15:08
That's a very good economic excuse. So, looking to the future, do you think CMOS remains the best platform for photonics or do you think there are other options that will eventually be better?

Chee Wei: 15:21
The booming of silicon photonics actually relies heavily on the CMOS compatibility. So I do not personally, I do not think there is a need to replace CMOS technology. We can still use it together with electronics and photonics. They complement each other, so you achieve a so-called better performance instead of replacing each other.

Don: 15:45
I agree, and CMOS technology is quite matured. We have very high yield and our cost of manufacturing is very low. So if we're able to implement that with silicon photonics, that will help us get mass adoption as well. Talking about mass adoption as generative AI takes off, data centers are facing escalating challenges. First of all, is talking about power consumption and the second, like what we mentioned, is data transmission speed. So I'd like to go into this. And, with data traffic exploding in recent years, what specific challenges do data centers face that silicon photonics can solve?

Chee Wei: 16:27
I think for silicon photonics. It is already solving some of the power consumption problem. Like, imagine if you are using everything on electronics. First, you can't transmit at such a high speed and second, the power consumption will going to make the data center explode. So by using silicon photonics it has already reduced maybe 50% of the power consumption if you are using all electronics. But of course, it's not enough yet. That's why I think we have to rely on other kinds of technology, such as some companies like Microsoft. They are actually immersing the whole data center under the sea, so this kind of cooling technology will still be required, besides the silicon photonics so latency is a big issue in data centers as we talk about data transmission.

Don: 17:26
How does silicon photonics help reduce latency? Because it's data. Silicon photonics is more for just data transmission from point to point b, but there is still the part where it fits into the memory, there's still the part that it fits into the CPU or the GPU. In that sense, do you think this is going to help?

Chee Wei: 17:44
There are actually a new kind of technology trend that people are working on in-package optical input output. This is something that is. One of the new things is the CPO that I mentioned just now. Another is this optical input output. It's basically some companies that are trying to implement optical input output in-touch or within the same package. So why they are doing this is because we know you can actually transmit a lot of signals very high speed and very high capacity. So they are trying to implement this in package. But this is something that is very new. Now it's still at a conceptual period. I believe maybe in the next 10 years we might be able to see something.

Don: 18:32
But for now it's still more for just data transmission and lower power consumption. What do you think the timeline looks like for silicon, photonics and EPIC to become mainstream in data centers, and are we talking years or decades, I think, since the first product commercialization in 2017, the percentage of silicon photonics used in data center transceiver is already increased every year.

Chee Wei: 19:00
So I believe in the next three or four years, you will actually overcome the competitors. Which is the 3.5 laser that people are using now. So it won't be a decade, it will be just within three or four years.

Don: 19:18
Let's take a look into how ready silicon photonics is today, focusing on core components such as the integrated laser, which is the source, the optical waveguide, which are the channels, and the modulators and demodulators that you talk about. And also we like to explore a little bit how silicon photonics can support advanced packaging by enabling different dies to be stacked together on a single chip, right? So, first of all, can you walk us through the main components that make up silicon photonics? How do all these pieces work together?

Chee Wei: 19:50
For silicon photonics die to function you require, of course, a signal source, such as a laser, and also a modulator to modulate the signals, and also a lot of power splitters and also multiplexer demultiplexer to split or combine the different wavelengths. So these are some of the very important building blocks in silicon photonics. And of course, as I mentioned before, the silicon photonics cannot generate light by itself. That's why we have to integrate other materials into the silicon photonics platform, such as the 3-5 semiconductor.

Don: 20:27
So optical waveguide is very often described as the backbone of silicon photonics. So what? Recent innovations have improved their performance and functionality.

Chee Wei: 20:37
So what recent innovation has improved their performance and functionality? I think one of the very common optical waveguides that most of us have seen is the fiber optics. That is one of the most generally seen optical waveguides. But for optical waveguides on silicon photonics, it'll be much smaller and you'll be like the width itself will be on the scale of hundreds of nanometers. The most important improvement needed for this kind of silicon waveguide is the propagation how to reduce the propagation loss, which I mentioned also in the previous question. The loss itself currently is still quite high and how to actually reduce the loss is one of the very basic and important challenges we have to face for silicon photonics.

Don: 21:32
So integrated lasers seem to be a big topic in silicon photonics, so can you explain why they're so important and what advancement are we seeing in this area, actually, for myself.

Chee Wei: 21:42
I've been working on these laser integrations since my time in A-Star in year 2007. So why this is very important. I mean, laser itself is a signal source. It's a source right. Just like you have to have a current source to be able to work for electronic IC, so this is the same for electronic IC. And one of the main challenges is the incompatibility of the materials. Like if you ask us silicon foundries, they are not going to make 3-5 materials for you in their silicon foundries, so it will require two different foundries working together. It could be in the same compound or different companies, but they have to be able to make these two compatible on the same way first. So this is something that is very challenging. And after you integrate them, how do you actually resolve the heat issue? Because for Photonic IC, one of the main thermal generators is actually the laser, not so much on the modulator or the splitter or detectors. The biggest part is actually coming from the laser source, so it will actually accumulate heat on the silicon photonic die. So how we actually resolve this issue is also a very important question.

Don: 23:08
So we're seeing some advancement in this area?

Chee Wei: 23:10
Yes, there are different companies that come up with a different solution, such as making a heat sink on the silicon die to actually propagate the heat from the laser to the substrate. But these are more like know-how for the certain companies. These are more like know-how for the certain companies, but I mean we need to have more solutions for this and also for silicon laser integration. Besides the wafer bonding kind of technology we still have to rely, I mean explore more different types of technologies such as packaging. How can we actually use packaging to solve this issue? And also, can we also use those kinds of mass transfer printing kind of technology to do the same thing? So these are different types of advanced technology people are exploring nowadays to do this laser integration.

Don: 24:04
Moving on. So modulators and demodulators are also very key for the fast data transfer that we're talking about. So how have they evolved recently? To support faster speed or more reliable connections. When silicon photonics actually hit the 200G patch pump for modulation per channel, it has actually almost reached a bottleneck. So to increase further, there are researchers and also companies that suggest to put lithium niobate material, doing something like the laser integration to integrate this material onto silicon photonics. So for lithium niobate the speed itself is actually much faster. The bandwidth is much higher than silicon. It can actually go up to like 100 gigahertz or 200 gigahertz just by implementing the lithium niobate material onto silicon. So in the future it won't be just a pure silicon material die. It might be a combination of different materials, such as the light source from 3,5, modulator from lithium niobate and also detector from germanium. So this is what I see the future will look like.

So it sounds like for us to integrate everything into one chip. We now start to see the trend that we need other materials rather than just silicon yeah, it's something like a silicon system on chip. So in terms of you talk about packaging right, advanced packaging and we have electronic die and we have photonics die into one single chip, one single package, can you talk a little bit more about that?

Chee Wei: 25:43
We are calling this, the co-packaged optics. So this is how we actually can stack together electronic IC maybe on top of a photonic IC. But of course for this, you require some other new fabrication techniques, such as how to make the TSV the true silicon via, to actually connect the electronics part from the top to the surface, add to the bottom and then connect to the same substrate. So this is something that we need to develop for this kind of advanced packaging and also for photonics part. You will have one tricky issue is how do you actually tap out the signal, and for that, usually when we do packaging, we will require alignment of a fiber optics to the silicon die.

And if we are going to do stacking of electronic IC onto the photonics IC and then this will actually bring another issue how do we actually align this fiber onto the silicon photonics die? Because most likely we'll align this fiber onto the silicon photonics die, because most likely we have to flip over the silicon photonics die in order to do the flip-trip bonding or the die attach, so we cannot see the waveguide anymore. So how do we actually align the fiber optics? So this is one of the very big issues and also, if you align after the many rounds of the die stacking, the fiber will actually drift, so the coupling efficiency will reduce. This is also another issue. Of course, I'm talking more details about the technology, but this is something that we have to. One is the TSV and another is how do we actually do the fiber alignment with this kind of under this kind of advanced packaging?

Don: 27:34
Yeah, so this is also what we see in the semiconductor industry. So, with our silicon getting to smaller nodes, we talk about five nanometer, three nanometer and then down to two. Right, having stacking two dies together in a single package is also very challenging right now as it gets smaller, because you have more pins, you have more balls and you try to stack it on top and press it down into one package. Any misalignment that means the whole package is gone. So, yeah, so we get a lot of that, but seems like adding silicon photonics onto it seems like adding silicon photonics onto the package itself is going to be even more challenging. From an advanced packaging perspective. Yeah, as fuel to a fire. So what are some of the biggest technical challenges that you see that still need to be overcome? Besides, when we talk about adding more materials the packaging side what are the technical challenges that still need to be overcome for silicon photonics to reach its full potential?

Chee Wei: 28:30
Of course, just like we talked about, one of the main issues is the loss on the silicon photonics chip and another is the integration of different materials to realize their functionalities and, of course, another to make it into full scale. I think the current issue for silicon photonics is that the market is still not fully matured yet. If you compare to electronics IC market, which is huge, electronics photonics IC is like a baby compared to an adult. So it will require a lot of companies working together to make more products based on silicon photonics. Then you will actually pick up and a lot of these kinds of problems, the technical problems. I believe it will be solved by then.

Don: 29:22
So we need the smart people, and a lot of smart people, to come in and work on this and to bring it to the next level.

Chee Wei: 29:27
Just like electronic IC. I believe there were a lot of issues, but because a lot of companies are pumping in money and people onto this technology, so it will become as virtual as today.

Don: 29:39
Exactly so, going on this train of thought, you know for silicon photonics, right, or EPIC, we talk about train of thought. Silicon Photonics or EPIC we talk about predominantly is on data centers right now, because of its lower power consumption and faster data transfer rate. So what other application are you seeing for Silicon, photonics or EPIC besides data centers?

Chee Wei: 30:00
Of course, data center or the optical communications are the two main applications for silicon photonics at the present, but silicon photonics itself is a platform technology. It can be implemented onto many areas, such as the optical sensor. In the past, I have worked in a company that actually working on using silicon photonics for water sensors, but it's actually at a very early stage, and another application is like doing optical computing. There are also some startups that are actually working on this and also just I mentioned the optical sensor, optical computing. They are also using silicon photonics for like biomedical purposes, like doing DNA sequencing. This is something that, of course, I'm not familiar with, but I actually know some companies are working on DNA sequencing using silicon photonics.

Don: 31:01
Sounds like it has a wide application. We'll be excited to see more about that. So, looking forward, what other emerging technologies or innovation in silicon photonics are you most excited about?

Chee Wei: 31:13
Personally, of course, I'm actually quite excited about this optical input-output, that we can actually connect the different chips in the same package by using optical signals. Of course, I believe it's very challenging, but it's something that I feel very excited to hear about this Because in this way I believe in the future maybe in our laptop or computer the speed will be even higher than now, which is actually limited quite a lot on this kind of electronic transmission.

Don: 31:46
So it seems like once we solve that problem, a lot of things will happen. In silicon photonics, we'll go past the barrier and go for faster speeds and lower power consumption. As we explore today, silicon photonics and EPIC are poised to redefine the future of data transmission, computing and much more. They're not just incremental improvements. They're representing a fundamental shift in how we think about technology, speed, efficiency, and integration. Chih-wai, it's been incredible hearing your insights on the breakthrough happening in this space and the exciting roads ahead. Thank you for helping us understand why this technology is on track to be the next big wave in semiconductors.

And with that, ladies and gentlemen, we wrap up today's episode of Advantest Talks Semi. A huge thank you to Dr. Lee Chee Wai for joining us and sharing his deep expertise. For our listeners, we hope this episode has shed some light on the transformative potential of silicon photonics and EPIC. If you enjoyed this conversation and want to stay updated on the latest in semiconductor innovation, don't forget to subscribe and share this episode. Thank you for listening. We'll catch you the next time.

Chee Wei Lee: 32:43
Thank you, Don. Thank you, everyone.