Thermoreflectance and Solving the Thermal Puzzle in 3D Packaging

Estimated reading time: 9 minutes

LaRont: Would you explain a bit more about your metrology technology?

Shakouri: Originally, we weren’t thinking about thermal, because it was already being measured by infrared emissions. We developed a technique that uses optical imaging using optical wavelengths of light. Light is a rainbow with different wavelengths. It can characterize different materials.

LaRont: At SEMICON two years ago, everyone was talking about the need to make gains in metrology because knowing what's happening inside those packages is increasingly difficult. At that stage, the result of failure is very expensive. Microsanj’s thermoreflectance addresses this concern directly. How is it groundbreaking?

Shakouri: Thermoreflectance is a fundamental physical phenomenon discovered more than 70 years ago. It says that if light hits a surface, it will be reflected based on the surface's temperature, because the material's index of refraction depends on its temperature.

This technique can achieve submicron spatial resolution, a 10x improvement over what has been traditionally achieved. Why is this possible? First, because it's visible. Second, you get different lights for different materials with different properties. You measure the colors of the rainbow. For example, near-IR light can go through silicon. So, you can actually go through a material and characterize it based on how it interacts with light, how it absorbs and reflects that material’s heat profile.

Because this falls under imaging, and cameras have become so powerful—your cell phone, for example—we have now developed an imaging technique that allows us to actually look at a semiconductor. Based on the light, you can now see the whole temperature profile on top of that circuit and how the energy is propagating through the package. That visualization has allowed us to look, fundamentally, at what is happening in the transistor, the interposer, and the TSVs. It’s quite amazing.

LaRont: Has your measurement work helped identify any thermal management solutions for these types of packages?  

Shakouri: Yes, as we study these packages, we have also been studying diamonds as a critical element and potential solution because of its very high thermal conductivity. My wife loves diamonds, but these are a different kind: an extremely powerful material for thermal conductivity. We discovered that if you can design and use a thin layer of diamond correctly, with the right interface, it can remove heat.

Stanford University, one of our customers, has fabricated diamond layers, measured thermal properties with our system, and will publish results in an upcoming webinar. Researchers have optimized the diamond's grain and placed it where the heat is generated to remove it. It’s basically a super-powerful, high-technology heat sink.

Remember, the worst thing that can happen inside a package is for the heat to spread everywhere. Using the diamond layer quickly conducts the heat resolves the problem.

LaRont: That is a revolutionary cooling technology. Processing the diamond material must be quite a process.

Shakouri: That is the uniqueness of their design. Our tool is used to characterize the boundary resistance, enabling them to optimize the material and interfaces at the boundary. The diamond by itself is fantastic, but if you don't maintain the quality of the interface, it won’t remove the heat from that level of the package. We leverage our thermoreflectance technique to measure the boundary resistance and heat capacity characterization of each layer of material, as well as to measure the whole system. We can measure the actual interface of the diamond layer all the way through the package.

We're still in the beginning stages of the diamond study. Material development, specifically interposer material development, will be a critical element.

LaRont: In your presentation, one slide (Figure 1) characterized several measurement techniques. Can you explain it?

Figure 1: Thermal analysis techniques temperature characteristics comparison.

Shakouri: This slide takes you through the different metrology methods currently used in industry, comparing them with thermoreflectance. Everyone in our industry knows that a thermocouple is a sensitive device used to measure surface temperature. It is a traditional metrology method that allows you to see down to 50 microns, if you're very good at it. However, it is a very slow way to measure.

Many in the industry also use infrared thermography, as I mentioned earlier in our discussion. It measures the heat emission. Infrared thermography has been a workhorse for the industry. For most materials, 3 to 10 microns is good enough, making infrared measurement perfectly suitable. Like using thermocouples, it is also a relatively slow process. In a millisecond, you can get maybe a microsecond's worth of transient to look at it. Up to about five to seven years ago, it met most of the semiconductor industry’s needs.

By contrast, the thermoreflectance technique has two components. You can see spatial resolution because light's wavelengths allow it, and you can measure down to a quarter of a micron. It's very fast, and the industry is pushing to make it even faster. The feedback we get is that our timing resolution is one of the most important factors. We achieve a 100 picosecond transient time but have characterized a 500 picosecond transient time. Why is this important? I can now measure so quickly that I can determine where the heat starts and how it propagates within the channel.

LaRont: What other aspects are important to mention?

Shakouri: Another important aspect of thermoreflectance is full-field imaging. I can measure the whole wafer. The ability to run a wafer or a whole panel through and measure it across the entire panel is making thermoreflectance even more valuable.

The last method in that list is micro-Raman, where you use a laser to probe the material and “look” at its properties. It is a single point of information: You characterize what is happening with lattice vibrations inside the semiconductor device by inferring from a single data point. It’s important, but micro-Raman does not measure the surface temperature and cannot be scaled. It's very good for scientific study, but not for commercial use.

For most people, what's important is running these tests themselves because they’re in production and need to move fast. It can’t require five PhDs to run the system on the production floor. Thermoreflectance can be integrated into a standard optical setup. In manufacturing, you always have a microscope to observe what is going on , i.e., optics, and now you can use it to measure thermoreflectance. Accessibility and ease of use are two other beautiful things about it.

We have optimized our tools so we can measure layer thicknesses from 10 nanometers, which is very thin, like the diamonds I mentioned, all the way up to 100 microns. I already know that our industry will say, “One hundred microns is not enough. I need to go deeper." We are currently conducting research to improve that metric.

Figure 2: Thermal imaging at different length scales.

LaRont: Using light and reflectance to measure accurately over a range of 10 nanometers to 100 microns is quite staggering. You also mentioned blind and buried via integrity in stacked microvias as a primary industry concern. Can thermoreflectance be used for that kind of structural-depth analysis as well?

Shakouri: Yes, for that use-case we have developed an optical pump probe Time-Division ThermoReflectance  (TDTR), meaning that we take a laser light and flash it on top of the surface. We have a camera. Vias are pathways for heat, and we can map the entire region of interest. We can actually see it. Unfortunately, most of our work is for specific customers, so I cannot share specific use case examples, but we can do full imaging of thousands of vias.

Figure 3: Laser thermoreflectance for structural depth analysis.

LaRont: In your paper, you mention a state-of-the-art parameter identification software. Is that a Microsanj product you have developed?

Shakouri: Yes. When you get millions of data sets, it can be very complicated to manage. We have developed a technique using AI to extract the information from those datasets to make a meaningful decision. Our software algorithm is our secret sauce. For example, if you measure a package stack and find a problem, how can you determine where it originated? We developed this software so we can go layer by layer in order and analyze them independently to discover where your thermal problem exists.

In another example, if your design calls for a 100-nanometer transducer in your package, but the manufacturer delivers 93, you could get a 50% error in your measurements. You have to be able to know that the transducer is 93 not 100.

Figure 4: A screenshot of Microsanj’s Parameter Identification software product.

LaRont: This sounds like the industry will benefit from this groundbreaking metrology work. How do standards play into it?

Shakouri: We are in the earliest stages of developing new techniques and standards that the industry must adopt to help all of us move forward. We are happy to work with the Association to establish a metrology standard, because they’re looking at how the industry can standardize.

LaRont: Well said. To summarize, with time-division thermoreflectance, you get accurate temperature measurements at a nanoscale level, in much shorter time durations, and with a wide temporal dynamic range; and you can analyze materials and thin films layer by layer in 3D heterogeneous integrated  packages..

Shakouri: Yes, and I want to emphasize that, with this technology, we can connect to a standard semiconductor product inline process, adding metrology so we can get a very good understanding from the beginning to the end. Having a tool that allows you to look at time variations from picoseconds to nanoseconds lets you go through the layers, go sub-micro, and deal with different materials. The industry needs this capability. It will allow us to mix and match and put together the best of the best to make these new AI microchips work better and more efficiently.

LaRont: Thank you for sharing your groundbreaking work. I will look forward to hearing more.

Shakouri: Nice to talk with you, Marcy.

• < Previous Page