The following transcript and video were originally presented at Thermal LIVE™ by Alex Guichard. For access and information on all Thermal LIVE events, both past and present, please visit https://thermal.live/.
Thank you, JJ, and thank you everyone for tuning in. Today I’m going to show you how Optoelectronics products are facing new thermal challenges, and how thermoelectric coolers in concert with smart package thermal design and integration can address those challenges. First, a brief word about the company I work for. Phononic is a thermal electric device and product manufacturer. Our headquarters and device fab is located in Durham, North Carolina. And we have sales offices as well as manufacturing partners of our end products throughout the US and Asia Pacific.
We’ve also been recognized as a CNBC Disruptor 50 Company for two years in a row. In this presentation, I’m going to first describe generally why and how thermoelectric coolers or TECs are used in Optoelectronics packages for cooling and temperature control. Next, I’ll talk about some trends we’ve noticed in laser package design with a focus on how those trends are creating new thermal challenges for Optoelectronics component designers and integrators.
And then finally, I’ll show some new technology examples of how we can tackle those new and increasingly stringent thermal challenges.
So, how do TECs keep lasers cool? Well, a thermoelectric cooler is essentially a solid state heat pump that can transfer heat away from a heat source to keep it cooler than ambient temperature. TECs consists of two parallel plates, typically made of a ceramic and an array of thermoelectric material pillars, we call elements.
Through the Peltier effect, TECs directly convert a current and voltage supply to heat flow. TECs can be built in a range of sizes and form factors, and are particularly well suited to cooling small components in constrained spaces where a vapor compression system simply is not feasible, hence their application to laser packages.
In most cool laser packaged designs, the laser diode assembly is a mounted directly to the TEC. And the TEC is not only the thermal management device for the light engine, but also acts as a sort of optical bench onto which other sensitive optics components are attached. When the laser is powered on, the TEC pumps the heat generated by the laser away from the assembly and into the package base. To control the temperature to a specific setpoint, a PID controller reads the temperature from a thermistor mounted near the laser and supplies the needed voltage and current to the TEC to maintain a temperature setpoint.
The TEC can also be powered in reverse bias and actually heat the laser to maintain operating setpoint in low temperature environments typically below 0C or so.
Now, you may be asking why lasers and optical networking components need TECs or any active cooling at all. At high temperatures, several characteristics of the laser degrade which impacts the transceivers a bit to maintain the specified signal strength and data rate. And suddenly, that 4K UHD stream of House of Cards becomes a choppy pixelated mess of a video feed.
First, laser wavelength shifts longer at higher temperatures. This shift can be large enough across the operating temperature range of the laser that it can actually cause major signal loss. This is especially true in applications like dense wavelength division multiplexing or DWDM, where multiple, closely spaced signal wavelengths are sent down the same fiber.
In this case, a TEC is required to maintain very tight control over the laser temperature and wavelength. Also, the maximum modulation frequency of a laser drops rapidly with increasing temperature, which means the rate at which it can transmit data is reduced. For most diode lasers, the maximum modulation frequency, that’s a mouthful, also called the 3dB cutoff frequency is the frequency at which the magnitude of intensity modulation drops to half of the low frequency value.
This cutoff frequency can drop by 10s of gigahertz across the laser operating temperature range, greatly impacting the rate of data transmission. Laser power conversion efficiency is also reduced, which means that more electrical power is required to maintain the same optical power. So for a given input power, signal strength, and therefore reach decrease with increasing operating ambient temperature.
All of these factors conspire to make temperature control very important for lasers used in optical communication, and TECs are really the only cooling technology that can provide the required cooling and temperature precision in these laser packages.
There’s a range of fiber optics products that utilize TECs for thermal control. TECs are integrated into many different types of laser packages, including TO Cans, Box TOSA, and Butterfly or also known as BTF packages. Those laser packages are then integrated into transceivers.
Transceiver applications that utilize cooled lasers include QSFP28, CFP4, and CFPA transceivers. Active cooling is almost universally required for tunable lasers or IPLA, as the TEC provides the wavelength tuning by taking advantage of the same behavior, I just mentioned on the previous slide, but in a controlled fashion, where changing the operating temperature of the laser tunes the laser wavelength.
In addition, various flavors of wavelength division multiplexing or WDM lasers, as well as the pump lasers for erbium doped fiber amplifiers, or EDFAs, often require active cooling as well. If you describe these products in terms of data rate and reach, we have observed that slower data rates like 10 gigabit per second or lower, usually only require active cooling for the longer reach applications, say in the 40 kilometer or more range. But higher data rate packages like 100 gigabit per second all the way up to 400 gigabit per second require cooling even for the shorter reaches down to, say, the five to 10 kilometer range.
This has led us to apply our own kind of general rule of thumb that the higher the data rate or the longer the reach, the more likely the application is to require active cooling with a TEC.
Now, I’ll shift gears to talk about how the laser light engines inside optical transceivers are evolving. The plot on the left shows how package heat density has dramatically increased with the ever increasing demand for bandwidth. So as the Netflix’s, YouTube’s, and Facebook’s of the world, drive the demand for more internet bandwidth, transceiver designs have shrunk very rapidly to fit more bandwidth per unit of rackspace.
And this leads to a dramatic increase in heat density at the laser package level. In working with our customers, we’ve noticed that the heat pumping requirement per unit area our customers are seeking has roughly increased by about 3x in the last two years alone. And we’re also designing for more and more components that must support the wider -40 to 85C operating temperature range, which is also called industrial temperature or I-Temp range.
In addition, power consumption limits set by industry wide, MSA specification impose strict power budgets on the transceiver and laser package. The TECs are one of the larger power consumers in a cooled laser package. It’s actually really important to consider heat pumping efficiency at multiple operating points when designing and selecting a TEC for a particular laser package.
All of these requirements are transpiring to pose a very challenging thermal environment for lasers. Looking forward to next generation package designs, lasers, detectors, and optics built directly into the board level will replace copper lines in a number of instances. And we expect novel packaging and cooling approaches will be required here. Also, non-hermetic laser packages are expected to reduce the cost and complexity of lasers, but introduce a number of cooling challenges mainly related to condensation in humid environments.
So, I’ve shown how the trends and optical transceiver and laser packaging products are introducing new and more difficult thermal challenges. And those three main challenges are; higher heat flux and temperature, reduced package power consumption, and new ways of integrating optics and networking systems.
Now, I want to show you some ways we can tackle these challenges through design and thermal modeling, innovative materials and processes. The processes and components via which the TECs are integrated into the lasers, and some new technology currently under development. But before going into gory detail about thermal design in these packages, I want to provide a general overview of a thermal electric system. The system can be very simple and consist of the laser assembly itself attached to the cold side of the TEC. We call that the Accept side, the TEC itself, and the heat reject system, which is typically in these cases the laser package base to which the TEC has been mounted.
The TEC in this thermal system is controlled with the driver circuit which uses temperature feedback to dial the TEC power and maintain a certain laser diode assembly temperature. A TEC operates most efficiently when the extraneous temperature Deltas on both the hot and cold sides of the TEC are minimized.
One of the most important ways we can address the thermal challenges presented by new laser packages is by properly defining the requirements during the TEC design process itself. So during this design process, our applications and thermal engineers work with laser package designers to fully account for all sources of heat and performance loss in the package.
In the diagram below, you can actually see some of the characteristics we must consider. The active heat load from the laser itself is the most obvious one, but if we take the package to be at a temperature of 85C, and the TEC cold side required to maintain a laser at 40C, some additional heat sources and losses will pretty quickly become apparent.
For instance, heat conduction through the wires that are connected from the laser assembly to the package sidewall and heat convections add to the overall heat load that the TEC needs to pump away from the laser. There are also additional parasitic temperature differences that are due to thermal resistance of the layers between the TEC and the laser. And those mean that the TEC has to get cooler than the laser temperature setpoint.
So, if the laser is being controlled to 40C, the cold side of the TEC may need to be at 35C or even lower, depending on the design of that laser diode sub-assembly that’s mounted onto the TEC. As mentioned on the previous slide, the size of this Delta is an important input to defining TEC operating point and efficiency. So, laser package designs should target low thermal resistances to minimize these parasitic delta Ts.
We take those aspects of the package design under consideration to create a first principles set of product requirements for every new application specific TEC design that we create. And you can take this even a step further where our customers have supplied their own package CAD to us. And we utilize that package CAD and create a full 3D thermal model that completely accounts for all of the additional heat sources and temperature gradients.
This allows us to make sure, one, we’re designing a TEC specific to this application to achieve the best efficiency possible right out of the gate. Two, it allows us to accurately predict TEC operating conditions and reduce the amount of time spent re-spinning designs. And then three, we’ve even actually provided insight into to the package designer, how they might reduce thermal parasitics, which in turn reduces overall package operating power even further.
The materials and processes that go into fabricating TECs can also have a big impact on laser package thermal properties. If you look at the anatomy of a thermoelectric cooler, there are several ways to improve performance, even while reducing both form factor and power consumption.
First, an ideal TEC uses a thermal electric material whose elements can be made incredibly thin and packed very tightly within the device to enable very high heat pumping densities, but not sacrifice efficiency. A high-performance TEC may also utilize thin ceramic substrate materials to reduce the thermal resistance of the TEC and increase the maximum DT achievable at a certain overall TEC thickness.
Third, a high-performance TEC requires extremely low contact resistance to the thermoelectric material elements. It turns out that electrical resistive losses at these interfaces end up limiting the performance of thin and small TECs. At Phononic, we have all of these design options at our disposal, and they along with our proprietary manufacturing process comprise our Pico-TEC device platform, which enables performance not previously available.
So, what does that mean in terms of real world TEC performance? Well, in relation to the challenges that I’ve highlighted earlier, increased heat pumping density, and a smaller form factor, and tighter power consumption requirements, we can offer advantages for both metrics.
From our research and from direct customer feedback, our TEC designs can provide up to a 60% improvement in heat pump and capability per unit area at about a 25% reduction in TEC power consumption. These two advantages can compound to provide big real estate savings inside the laser package as well as reduce the overall laser package power consumption in the most thermally demanding applications.
Now, TEC and laser package design are important to ensuring optimal performance, but the processes and components that the TEC is integrated with can have just as much impact. So like all thermal interfaces, the quality and thickness of the interface between the TEC and the laser package base can have a big influence on TEC performance.
For instance, if you choose the solder or thermal epoxy, the reflow or cure temperatures and overall package processing thermal budget can affect the thermal conductivity, percent voiding, and thickness, which directly defines the thermal resistance of that interface and can greatly increase both the TEC to package delta T and the maximum amount of heat pumping that is possible within the package.
So, characterization tools like X-ray or CSAM, also known as Confocal Scanning Acoustic Microscopy, can be used to inspect these varied interfaces and provide nondestructive feedback on the interfacial quality and how it may change throughout the laser package assembly process.
This slide shows CSAM scans of a TEC to package thermal interfaces. The top line of images shows how improper choice of solder that cannot withstand extreme temperature processing can degrade the thermal performance, and degrade that thermal interface. The TEC controller and driver used in the next higher assembly, which is called a Transmit Optical Sub-Assembly or TOSA, can also have a large impact on overall TEC and package power consumption, especially if the wrong driver is chosen that does not align well with the TEC operating point.
There are several TEC driver solutions available in the market today, and I’ve shown some examples from analog devices, maxim and linear technology here on the slide. Like most driver circuits, the efficiency drops rapidly at the low end of the operating range. So you definitely want to stay in the higher current range of this operating efficiency curve.
But at the same time, the efficiency increases with higher TEC load or higher TEC V over I. So, we recommend that the expected typical TEC operating point is actually used to help select the best TEC driver for the package, or vice versa. Use those efficiency curves to help inform the design of the TEC for your package. Also, we recommend PID control instead of PWM to maximize TEC heat pumping efficiency.
We are also working on a couple of technologies to address the next generation of optical networking components. One TEC platform under development enables cooled non-hermetic packages. This moisture resilient thermal electric device avoids corrosion issues associated with condensation. However, it’s still recommended for low humidity environments only and careful package design must allow for a sort of breathable exchange of air and moisture with the ambient environment.
If the TEC cools the laser assembly below the dew point of the environment, there’s a high risk of condensation on the laser and optics themselves, which will likely interfere with fiber alignment and lead to optical coupling losses. In fact, at a typical data center and humidity laser operating temperatures, it is highly likely that the cold side of the TEC will condense water out of the environment. So beware of the dew point of your customers’ operating environment when designing a cool non-hermetic package.
Another potential technology opportunity is a more tightly integrated TEC and laser package. This is particularly advantageous to TO Can laser packages. Cool TO Can packages are very small. A TO-56 header typically has less than a four millimeter diameter area to integrate all of the components necessary for a high speed modulated laser.
As such, these packages are incredibly space constrained, so any approach that reduces the overall real estate required for cooling can help the package designer integrate additional optics, modulators, or even photonic ICs that previously just couldn’t fit inside the package. Conversely, this approach can pack even more cooling into the same space to support extreme thermal requirements. In addition, this technology translates to fewer process steps and a simpler bomb for the package manufacturer.
So to summarize what we’ve talked about today, increased data rates and reduce package footprints are conspiring to create an increasingly challenging thermal environment for lasers and optical communications. Thermal electric coolers have been used in this space for decades and are still really the only technology that can achieve the stringent cooling requirements of fiber optic laser packages.
However, Phononics design approach and high heat pumping density, low power consumption TECs offer advantages that align well with these tougher thermal challenges. Also, package integration choices such as the TEC driver chip, solder selection, and process thermal budget should be made very carefully to achieve optimal TEC performance in the end product.
With that, I’d like to thank Thermal LIVE and the audience for the opportunity to talk today. Thank you.
Excellent, Alex. Great presentation. Lots of coverage. And I appreciate the background and the primer set that you provided as well. And I’m sure some of the audience wasn’t necessarily familiar with this technology to begin with. But we will be taking questions now. So if anyone would like to throw up any of the questions, they have anything they’ve puzzled over, we even had a few of the audience members ask questions before the presentation even began.
So I think a lot of people are really excited to hear from you about this, Alex. We do have a lot of questions, but we also have a lot of time to cover them. So that’s great. I’ll go ahead and hit up the first question. This one is about the maximum operating temperature of TECs and if there is a really high temperature, availability, a high max temperature device availability. Someone suggested in their question around 125 degrees C of continuous operating temperature.
So that’s a question about the package ambient, or the TEC ambient that it’s working in?
Yeah, it’s ambient and I’m assuming its max operating temperature as well. That there are product lines that can do that.
Yeah so we typically spec our TECs to operate in a 85C max ambient, but we encourage our customers to push the boundaries and 125C is definitely… We test to the in storage and in operating conditions. So, it’s definitely doable. We’ve got testing beyond 2000 hours and power cycling, for many 1000s of cycles at that temperature range. Our spec sheets say 85C, but even with the current technology platform that we’re working with, depending on the application, 125C may be doable.
Excellent. Thank you for illuminating on that one, Alex. Now, the next question, and I believe we covered some of this in your presentation, but I think some of the folks on the webcast would appreciate just illuminate a little bit more, it’s the value of a TEC in between the laser and the package. So the question revolves around the benefits of having the TEC there or just having the laser plate directly against the packaging [inaudible 00:26:38] and safe for suggestion if the packaging they were cold plate. So instead of using a TEC, what would get the best performance and other reasons why you go to a TEC over doing just that simple passive setup?
Yeah, so I think the best way to answer that question is really to talk about temperature. The lag time and the temperature variations. So you would need a MV space constraints that there are in these packages. So these laser packages are all totaled, maybe 10 to 20 millimeters on a side. And only a couple of millimeters thick. And then they go into this transceiver assembly that is, it’s basically eight by I think about 20 millimeters in cross sectional area, and only about 70 millimeters long.
And then they’re densely packed into a rack unit to provide very high bandwidth density in a data center or in a ISP location to transmit and receive data at incredibly fast rates. So, one is space constraint, there just really isn’t room to provide coal blocks or water cooling to each of these transceivers. And two, these lasers require essentially instantaneous, or response to fluctuations in the heat load. And that’s what the thermal electric devices can provide.
Excellent, so a lot faster response to the heat flux than a passive technique? Just a-
Yeah, faster response and much more aligned with the space constraints that these products have to deal with.
Excellent, thank you. I hope you answer that person’s question. They seem pretty excited to hear about it. The next one has to do, I’m assuming with the gaseous environment around a TEC. The person suggested that a lot of data sheets on TECs are generated in either nitrogen or vacuum. So that’s how their first cues are generated. And how can you translate information taken under those circumstances to for example, a CO2 environment by comparison? And the questions around the first and second cues and how those are generated, and if there’s a way you can communicate that from one gaseous medium to another.
Yeah, yes. So we actually… I don’t know how specifically CO2 compares to vacuum or nitrogen. When we do the performance modeling for our customers, we actually model the gas that’s inside the package as part of that model. So it’s basically taken into account as a parallel thermal resistance. And that’s how we actually get that built into our model.
So if someone were to reach out to you, you’d be able to provide them with information for their specific environment?
Absolutely, yeah. Yeah, in fact we frequently have customers ask to compare the performance of the TEC and these are, like I said, hermetically sealed, and it’s in an inert environment, but they’ll ask us to compare Xenon, argon, nitrogen. They want us to provide the different operating points for those different ambient gasses.
Excellent. So, in terms of solder materials, and someone who’s asking, what are the recommended solder materials, types of solder or conditions that you might need to attach the TEC to the plate and the laser?
Yeah, good question. It’s really dependent on your own package requirements, and really what you’re doing downstream of the TEC integration step. We’ve seen customers use… I will mention that an important requirement of some customers is no out gassing of whatever medium they use. So they try to shy away from fluxes. So fluxless basement 10 preforms is the one we see pretty commonly. we see sack 305 as a common attached solder used for these type of devices.
We actually also see thermal epoxies used pretty commonly. And if thermal performance and tech, power consumption performance is really important. We tend to try to steer customers more towards the solders than the thermal epoxies, just due to the better thermal interface that can be created. But yeah, those are three really common examples that we see out in the field used by our customers.
Excellent, excellent. Onto the next question, this one is asking if there are FloTherm dot PML smart part files for these devices, do you provide those?
No, we don’t. So we have a proprietary model that we’ve developed in house that we’ve kind of built a hybrid modeling platform that that interacts with FloTherm and some of the other modeling tools that we have out there. So we don’t provide that model. But we are very happy to work with customers to generate the modeling that is needed under NDA Of course.
So which might be a better answer than, “Yeah, we just have generic models.”
That’s right. And really the purpose of that is to, it kind of goes into our design process where we want to make sure the right design, and from a sizing perspective, the right device is used, and we want to help the customer with that. So that’s how we typically will run our modeling efforts, we’ll do it in conjunction and in partnership with the customer.
Actually, that is a great point, because the next couple of questions are about that design process. And someone else asked about how it actually helps with shortening the thermal design guidelines, and if you can provide some examples. So I think there’s a lot of interest in exactly how you guys at Phononics help with that.
Yeah. So I would say this, and I’ll provide the general and then the specific examples. When we approach a customer or a customer approaches us, we can ask for a specific set of requirements, including heat load and things like that. But we tend to back it up, and say, “Alright, hold on let’s get the active heat load from the laser. Then you can tell me how many wire bonds are to the cold side of your device? And how long are they? What diameter wire? What’s the material?”
We go into very in depth conversations. And that’s kind of that first principles approach that I was talking about where we want to hear, all right, what’s your package design like right now? And what are you trying to accomplish, and we’ll get you a design that works best? And a perfect example of shortening that design cycle is, we had a customer approach us that said, “Look, we bought these devices from a competitor. And we’re getting thermal runaway issues where they’re not able to cool sufficiently for the application, before the hot side, temperature starts taking off.”
And what we found in working with them, and this is a great example of where them supplying us their package CAD was really helpful. We did a full thermal analysis for them and found that there were some heat loads that hadn’t really properly been accounted for. And so the TEC itself was severely undersized for the application in terms of heat pumping capacity. And so I think in that case, the shortening of the design cycle was… We would have eliminated that first unsuccessful loop, had we been able to do that in depth design a little more early in the process, instead of them just going out and buying TECs.
So, yeah, I think it’s a little bit of making sure we ask the right requirements questions at the outset. And a lot of design due diligence to make sure we’re accounting for everything in the package.
Excellent. So this questions kind of about what you end up giving to a customer so that they can maybe verify in their simulations if the TEC is a good solution. I’m wondering if you don’t provide the FloTherm .pdml smart parts, do you provide Q-max information, T-max information, V-curves? That kind of stuff. What would you end up delivering to a customer after you guys have your evaluation?
Yeah, so what we typically deliver is a full design proposal that can include some performance modeling tools that we provide, this is typically not a FloTherm model, but we can provide some calculators for different operating points and things like that. But also at the very beginning, when we are talking with the customers, we’re asking things like, “Alright, what is your laser diode assembly temperature? What are all of the layers in between the laser and the TEC? How are you attaching it?”
We ask a ton of questions to then with that design proposal, we deliver a performance model for the operating points of interest that the customer intends to run at. And in particular, those operating points are usually the boundary conditions like max operating temperature range, and kind of the coolest, the laser will need to be cooled in the application.
So we provide those boundary conditions and then a more typical operating condition. And so at that proposal stage, lots of performance information, and we can work with the customer to provide additional models that are a little more free form as well.
Thank you, Alex. Excellent. Now, just to give you a heads up, there are a lot of questions, and some are still coming in. So we’re going to try to go through as many as we can, before time’s up. We do only have a few more minutes. So I do want to let you know that before we dig in and these are great answers. Great responses, and I’m sure the audience is pleased.
Nope, no problem.
To the next-
And I just wanted to let the attendees know that I’m happy to follow up via email afterwards to provide some more answers if they don’t get answered in the time allowed.
Thank you. Excellent Alex. The next question is about… And I think you already touched on this a little bit when you answered the solder question. Is what thermal interfaces are most important in these packages?
Yeah, basically on either sides of the TEC, that cold side interface is important, because it can be one it can be very definitive in terms of heat spreading and the amount of delta T between the laser and the TEC. And I mentioned in the slide but not voting verbally, thermal shorting, if there are any components, large surface area components that directly bridge the hot pack and sidewall to the TEC cold side, that’s a basically a thermal shunt resistance.
So understanding those interfaces, both in one dimension in two dimensions is really incredibly important. And that hot side interface is also incredibly important because if that hot side interface is badly formed, if there’s a high degree of voiding, you’ll end up with TEC hotspots that could not just impact performance, but could impact long term reliability. And also, if it’s too thick, you’re not going to get the expected performance because your hot side will be getting too hot. And your delta T will need to be larger than you had previously modeled.
So really, those hot and cold side interfaces are the most important and the ones you want to make sure you engineer very carefully.
Great. Now this has some relevance to that. I think it’s a little bit more of a basic question involved in that. And that’s why is minimizing delta T in the package important?
That’s a very great question. And it’s actually it is related to how TECs work. So unlike other heat pumps, say vapor-compression systems, the COP of a TEC is very dependent on the delta T across it. So at a hot side temperature, the lower the delta T across the TEC, the higher the cooling COP will be. So you want to make sure you get that, if you’re if you’re designing for a 30C delta T application, meaning 30C between ambient and the laser, you want to make sure that that TEC is as the delta T across the TEC because it’s going to be larger due to those thermal resistances.
You want to make sure that that TEC delta T is as close to 30C as you can get. And it’s just like I said, just due to the dependence of COP and how it increases with decreasing delta T.
All right. Excellent. Now, we only have a couple minutes left. But I think we can go over a little bit, we have a ton of questions left as well. And they’re all really good questions. I’m also very interested in hearing these answers. But we’ll just go to the next one. And if we can’t take the rest, we’ll talk about that a bit. The question is, why does PID control maximize cooling efficiency? The person who asked the question says, “I thought the purpose was to control the temperature within a narrow range, so what is the meaning of efficiency in this context?”
That is also a good question. And it’s actually again related to the behavior of COP for thermo electric devices. So if you were to plot the COP curve as a function of voltage or current of TEC voltage or TEC current, you would find that it’s got a pretty well pronounced peak at a slightly lower voltage points. Well below its maximum rating. So, what you want to be able to do is operate that TEC in both a high heat pumping mode and a high efficiency mode.
So you could conceivably size the TEC to operate at that high efficiency mode only at its max PMW voltage, but you will sacrifice that higher heat pumping capability for those rare extreme temperature conditions that you need to support. So it’s really about properly sizing the thermal system to be able to allow the TEC to operate at both that high heat pumping mode or high delta T mode, and a more kind of low power maintenance, high efficiency mode.
And the PID controls that are used are plenty good enough in these cases to provide very, very tight temperature control in the application here.