Hot Trends In Semiconductor Thermal Management

Cooler bonding, microfluidics, and engineered TIMs help get the heat out.


Increasing thermal challenges, as the industry moves into 3D packaging and continues to scale digital logic, are pushing the limits of R&D.

The basic physics of having too much heat trapped in too small a space is leading to tangible problems, like consumer products that are too hot to hold. Far worse, however, is the loss of power and reliability, as overheated DRAM has to continually refresh and chips become even more stressed in high-heat sectors such as automotive.

“In an ideal world, you would have your die made of copper and your substrate would be 100% copper,” said Nathan Whitchurch, senior staff mechanical engineer, at Amkor. “But even if you could, you wouldn’t get any more performance because of some other limiting factors in the package.”

Thermal issues are becoming an earlier design and packaging decision in 2.5D and 3D packages. “Thermal dissipation is one of the key issues that we have to consider, in memory on logic, but also in logic on logic stacking, said Yin Chang, senior vice president of sales and marketing at ASE.

As the industry seeks solutions, microfluidics and thermal interface materials (TIMs) are key areas of development. The former is seeing breakthroughs. The latter is making incremental improvements. To remove heat, liquid coolers can be direct bonded to chips or channels can be built into the chips themselves. On the TIM side, sintered silver epoxy is gaining use.

Microfluidics may soon make the transition to production. “I’m betting that microfluidics is going to start appearing beyond just the hyper exotic places, especially if you start stacking high performance logic,” said Rob Aitken, distinguished architect at Synopsys. “If you don’t do anything about cooling, then your stacked logic is limited to the thermal dissipation that a single die had. There’s a massive economic push to solve such problems. Given that, and given people’s creativity, I would bet that somebody will solve it in some clever way.”

Status of microfluidics
For the last 40 years, commercial microfluidics technology has been just around the corner. The idea of embedding liquids in micro/nanoscaled channels to cool semiconductors was first described in a now-classic paper by Tuckerman and Pease in 1981.1 Variations have been tried ever since, and now some projects are showing real and practical promise for cooling.

Two years ago, a group from École Polytechnique Fédérale de Lausanne (EPFL) in Switzerland demonstrated a prototype that brought the cooling liquid as close as possible to the heat source. The design was a working version of an often-discussed microfluidics goal — integrating channels directly within a chip, rather than relying on TIMs or bonding, the latter of which makes the commercial market skittish due to reliability issues.

The Swiss team overcame the challenges by creating, in the language of their paper, “a monolithically integrated manifold microchannel cooling structure [in which] heat fluxes exceeding 1.7 kilowatts per square centimetre can be extracted using only 0.57 watts per square centimetre of pumping power.”2

The paper attracted interest from investors, and their idea progressed from the lab into a start-up. Its first author, Remco van Erp, along with his EPFL professor Elison Matioli and COO Sam Harrison, have co-founded a company called Corintis,3 which has received funding from the Swiss government to create commercial versions of their innovation.

“From the thermal point of view, Corintis’ approach is a very interesting cooling solution, as the coolant can get as close as possible to the location of the heat sources and several thermal barriers can be eliminated in this configuration,” said Herman Oprins, principal member of technical staff at imec. But he cautioned that commercial adoption is not a given. “This is a disruptive cooling solution that requires a tight co-design between the fluidic channel structures and the electronic devices in order to achieve the full potential of this cooling method. It is very well suited for challenging applications with very high power densities, such as the power bar structures shown in the paper. For CMOS applications, with power densities in the range of several hundreds W/cm², separate cooling blocks with more relaxed channel diameters of several hundreds of µm could be used.”

Imec presented its own microfluidics prototype three years ago. Its press release describes the concept as “a Si microchannel heat sink assembled to a high performance chip for cooling the latter one. [It] achieves a low total thermal resistance of 0.34K/W to 0.28K/W at less than 2W pump power.”4

“We have two main types of prototypes,” Oprins explained. “One is the silicon microchannel cooler. There, the main development is the bonding to the chip with low thermal resistance. The second one is the direct liquid cooling on the chip using 3D printed cooling geometries in complex shapes.”

While the imec efforts have not yet been commercialized, companies already are offering similar designs, according to Oprins.

Describing the origin of the imec prototype, Oprins said, “We leveraged our knowledge of wafer-to-wafer bonding to bond the cooler to the chip, with a very low thermal resistance of less than 1 mm2-K/W. Thus, instead of using a thermal interface material, we can use fusion bonding, or oxide bonding, or metal bonding. The main advantage for semiconductor processing is that there could be strict tolerances with a very thin line.”

Oprins points to several issues. “For the mechanical integrity of your package, you need to compensate for the absence of the lid with a stiffener ring,” he said. “If you make the channels too small, the pressure drop that you need to push your coolant through will be too high. You’re limited how small you can go with the liquids.”

However, he noted that while higher pressure is a potential drawback, it is not a show-stopper. “The main reasons for slow adoption are reliability concerns (leaks), the need for maintenance, and system complexity.”

Fig. 1. Various cooling approaches. Source: Imec

Fig. 1. Various cooling approaches. Source: Imec

Oprins classified the current and proposed commercial approaches to liquid cooling into four distinct types:

  • Bolt-on cooler. This is the current state-of-the-art in data centers. A cold plate sits on top of the lid instead of a heatsink. TIMs are used above and below.
  • Directly bonded cooler. This configuration is beginning to be adopted in some places. The cooler is directly bonded to the chip, with only one layer of thermal interface material. Imec’s prototype uses this layout, with modifications.
  • Backside cooling. Only presented in research, this layout allows the coolant to be closer to the heat source. Instead of bonding, it uses a dielectric liquid that makes direct contact to the chip. Because there is a vertical connection between the liquid and the chip, it avoids the thermal gradient problems of a lateral design.
  • In-chip cooling. This is the idea that Corintis is trying to commercialize. The coolant is contained within channels embedded in the chip. While it can offer optimal cooling, one potential challenge is that there may not be enough space for channels at lower pitches.

In addition to this work, Sam Sadri, senior process engineer at QP Technologies, recently showed off a prototype of an internally cooled package. Created with 3D technology, it’s made of ceramic alumina that uses thick-film technology for top metallization, onto which several SiC FETs will be attached.

“Alumina is already an oxide (Al2O3) and copper oxidizes easily, so the two oxides bond together, and that’s how this interface is made,” Sadri explained. “This is by far the cheapest way to build a power module, with a ceramic. There are ways to further reduce the cost. An isolated metal substrate (IMS) basically is like any PCB fabrication technology, but it uses heavy copper.  While most PCB copper contains 0.25 to 0.5 ounces of copper, this is closer to 3 or 4 ounces. This is what I’ve seen as far as something that’s more cost effective than alumina with the same footprint.”

The prototype’s dimensions are approximately 4″ x 2 ½” x ¾” deep. While it’s thicker than a typical substrate, what makes this rectangular structure special is that it has channels that tunnel all the way through, with exit holes on its shorter sides. “This is one of the coolest things I’ve seen in power,” said Sadri. “When you power it up to full duty cycle, the module puts out a lot of heat. How do you get rid of the heat? You send a coolant such as cold air, nitrogen, coolant, or some other cold substance through the channels. As it is operating, it’s also cooling down, too.”

Improving TIMs
As shown above, both bolt-on coolers and directly bonded coolers use TIMs to optimize heat conduction between the chip and cooler, as do many other configurations. TIMs use a wide variety of materials that can include “thermal greases, gap fillers, insulating hardware materials, thermal pads and films, graphite pads and films, thermal tapes, phase change materials and thermal epoxies [as well as] thermally conductive ceramics, e.g., aluminum oxide, aluminum nitride and beryllium oxide,” according to a recent review of cooling systems.5

Yet many TIMs turn out not to be as efficient as their widespread use would suggest. “Thermal interface materials become an important thermal bottleneck as the liquid cooling performance improves,” said Oprins. “The system integrators have a lot of questions about how TIMs can be replaced by better performing materials and what the reliability risks are.”

The challenge is to discover a material that has very high thermal conductivity, while at the same time being very pliant and soft so it can follow the topology of different components.

“Typically, most materials that have good conductivity are also very rigid, so not only will they not conform, they can add to the stresses,” Oprins explained. “You’re looking for a combination that’s difficult to find. So there will be no single material that will have these properties. Researchers will have to engineer one by making composites. For example, instead of just using a silicone paste like in the past, now there can be thermally conductive particles inside to increase the thermal performance. There can be composites. There even could be carbon nanotubes or graphene sheets. There’s a lot of advancement in that particular area. We started with the silicone based materials, and eventually we will end up with metal-based thermal interface materials, but there are a lot of reliability issues to be solved first.”

Given the urgent need for novel materials, Amkor’s Whitchurch emphasized that all engineers should respect how crucial material science breakthroughs will be to solving thermal problems — and that the industry has a long way to go to find materials that can be flexible, reliable, and economical.

We’re exploring many different TIMs, which are no longer polymer-based,” he said. “Things that used to be exotic are becoming less so, like the sintered silver category, and you end up with a very hard, high thermal conductivity matrix of a silver alloy between lid and die. Another example would be softer metal materials, such as ones based on indium. Gallium scares people, because it reacts with aluminum, so we haven’t seen as much of that kind of environment. A couple years ago, we were talking frequently about phase change material, but that seems to have died off as people realized the reliability and other advantages just weren’t there. The other things that I’ve seen like graphite pads, they also have some engineering challenges that are too difficult to overcome. Graphite in a single direction is highly thermally conductive, but actually getting that into a package is a difficult challenge.”

To get rid of the power in a flip chip package, Sadri said, “Traditionally, a back side metallized SiC power FET die is attached to a heatsink using solder (e.g. AuSn). Today, sintered silver epoxy has shown better thermals, so folks use both pressureless (e.g. Atrox) or pressured sintered epoxy (Argomax). In a flip chip scenario, a heatsink is designed with nickel-plated copper which kisses the back of the chip with thermal interface material (TIM) in the interface. Other innovations use a number of wires in the back of the chip and then land the wires on a ground plane in the PCB to improve thermals. Copper-tungsten and copper-moly are other types of heat sinks that folks like for the CTE match to silicon, but they are expensive. Copper is still the best thermal interface and very cost effective.”

An alternative approach would eliminate the need for TIMs altogether, which is one of the motivations behind imec’s work on microfluidics. “You want to find alternative cooling solutions so you can avoid the interface materials, so that’s what we’re doing with the liquid cooling,” said Oprins. “We want to bring it closer to the chip so we can eliminate these materials. I would say that’s the bottom line. You either improve the materials or you get rid of them.”

The result of these challenges is that solving thermal problems has increasingly moved up the list of budget priorities. “Customers are often surprised that they have to dedicate so much budget to thermals,” Whitchurch said. “But in order for a package to simply work, we’re all going to have to pay attention, because ultimately a package that works is cheaper than a package that doesn’t work. We’re starting to see a lot of our customers come to those realizations and start to engage our more advanced engineering, skill, and experience for products that 10 years ago would never have shown up on my radar.” 

Still, change takes time. “The industry is very conservative,” Oprins said. ”It takes a lot of convincing to switch to something they don’t know. Everything you introduce comes with a lot of complexity. I understand the reluctance to adopt something new until it’s proven to have worked and all the liability issues have been tackled. Nevertheless, there are many great ideas out there. We know they still need a lot of work, and we are looking for new recruits who can help.”


  1. D. B. Tuckerman, R.F.W. Pease, High-performance Heat Sinking for VLSI, IEEE Electron Device Letters 2,5 (1981)
  2. Remco van Erp, et al., Co-designing Electronics with Microfluidics for More Sustainable Cooling, Nature 585, 211–216 (2020)
  3. News release: Corintis wins CHF 150,000 to disrupt cooling technology and enable the next generation of computing,
  4. News release: A miniature microfluidics heat sink for high-performance chip cooling,
  5. Górecki, K.; Posobkiewicz, K., Cooling Systems of Power Semiconductor Devices — A Review. Energies 2022, 15, 4566.


Denis McCarthy says:

When considering heat in electronics the designers need to evaluate the entire package. While working for a laminate manufacturer and now that I am in PCB manufacturing again there are some very good alternatives to FR4 that multilayer structures can be built with. These thermal materials can be used in so many ways to combat the heat or work in conjunction with heavy coppers, heat sinks, or other technology!

Leave a Reply

(Note: This name will be displayed publicly)