A Solution to Heat Dissipation of 3D Packaged Chips

As the industry moves to 3D packaging and continues to scale digital logic, increasing thermal challenges are pushing the limits of research and development. Too much heat in a small space can cause real problems, such as products being too hot to hold. Overheating DRAM needs to be constantly refreshed due to power and reliability losses, making the chip even more stressed in high-temperature industries such as automotive.

In ideal conditions, the chip would be made of copper and the substrate would be 100% copper. Even if that were possible, the chip wouldn't gain more performance because of some other limiting factor in the package.

Thermal concerns are becoming an early design and packaging decision in 2.5D and 3D packages. Heat dissipation is one of the key issues that must be considered, both in terms of memory and logic on the logic stack.

In search of solutions, microfluidics and thermal interface materials (TIMs) emerged as key areas of development. To dissipate heat, a liquid cooler can be attached directly to the chip, or channels can be built into the chip itself. On the TIM side, sintered silver epoxy is used.

Microfluidics may soon transition to production. It's going to start showing up in super exotic places, especially if you start doing stacked high-performance logic. If no cooling measures are taken, the stack logic will be limited to the heat dissipation of a single chip. There is a huge economic impetus to address these issues.

For the past 40 years, commercial microfluidics have been within reach. The idea of embedding liquids in micro/nanoscale channels to cool semiconductors was first described in a now-classic paper. Various variants have been tried since 1981, and now some projects show real and practical cooling promise. From a thermal point of view, this is a very interesting cooling solution, since the coolant can be located as close as possible to the heat source, and several thermal barriers are eliminated in this configuration. A working version of the microfluidic target integrates channels directly into the chip, rather than relying on TIMs or bonding. The latter has destabilized the commercial market due to reliability issues. This is a disruptive cooling solution that requires tight co-design between fluid channel structures and electronics to realize the full potential of this cooling method. It is ideal for challenging applications with extremely high power densities. For CMOS applications with power densities in the hundreds of W/cm² range, separate cooling blocks with more relaxed channel diameters of several hundred µm can be used.

There are two main types of archetypes. One is a silicon microchannel cooler, and the main development is the bonding to the chip with low thermal resistance. The other is direct liquid cooling on the chip using complex-shaped 3D-printed cooling geometries.

The cooler is bonded to the chip using knowledge of wafer-to-wafer bonding with very low thermal resistance of less than 1 mm 2 -K/W. Use fusion bonding, oxide bonding, or metal bonding instead of thermal interface materials. A major advantage of semiconductor processing is that very fine lines can be made to tight tolerances.

A reinforcing ring is often required to compensate for the absence of a lid for the mechanical integrity of the package. If the passages are too small, the pressure drop pushing the coolant through will be too high. And the volume of liquid is finite. The main reasons for slow adoption are reliability issues (leakage), maintenance requirements, and system complexity. Higher pressure is a potential disadvantage, but it's not a hindrance.

Divide commercial liquid cooling methods into four distinct types.

Bolt-on coolers, the most advanced technology available in data centers. The cooling plate sits on top of the lid instead of the radiator. TIM is used above and below.

The cooler is bonded directly to the chip with only one layer of thermal interface material. Some places are starting to adopt this configuration.

Backside cooling This layout has only been proposed in studies and allows the coolant to be closer to the heat source. Instead of bonding, it uses a dielectric fluid that is in direct contact with the chip. There is a vertical connection between the liquid and the chip. Therefore, thermal gradient problems of lateral designs are avoided.

On-chip cooling involves the inclusion of coolant within channels embedded in the chip. While it provides the best cooling, one potential challenge is that there may not be enough room for lower-pitch channels.

In addition, a prototype of an internally cooled package, created with 3D technology and made of ceramic alumina. It uses thick film technology for top metallization and multiple SiC FETs will be connected to it. Aluminum oxide is already an oxide, and copper oxidizes easily, so the combination of these two oxides is how this interface is formed. This is currently the cheapest way to manufacture power modules from ceramics, reducing costs. Isolated Metal Substrate (IMS) is basically like any PCB manufacturing technology, but it uses heavy copper. While most PCB copper contains 0.25 to 0.5 ounces of copper, it's closer to 3 or 4 ounces. This is something more cost effective than alumina with the same footprint.

Although the dimensions of the prototype are thicker than typical substrates, what makes this rectangular structure special is that it has channels running through it, with exit holes on its shorter sides. When powered up to full duty cycle, the module dissipates a lot of heat. How to get rid of the heat? A coolant is sent through the channel, such as cold air, nitrogen coolant, or some other cold substance. As the coolant runs, it cools down.

Both the bolt-on cooler and the direct bond cooler described above use a TIM to optimize heat transfer between the chip and the cooler, as do many other configurations. TIMs use a variety of materials including thermal grease, gap fillers, insulating hardware materials, phase change materials, and thermally conductive epoxies such as aluminum oxide, aluminum nitride, and beryllium oxide.

However, it turns out that many TIMs are not as efficient as their widespread use would suggest. As liquid cooling performance increases, thermal interface materials become an important thermal bottleneck. System integrators have many questions about how to replace TIMs with better performing materials and what the reliability risks are.

The challenge was to find a material that has very high thermal conductivity and at the same time is so flexible that it can follow the topology of different components.

In general, most materials with good electrical conductivity are also very stiff, so not only will they not adapt, but they will add stress. No single material has these properties, so one had to be engineered by making composites. Thermally conductive particles can now be added inside to improve thermal conductivity. There can be composite materials. There might even be carbon nanotubes or graphene sheets. There's been a lot of progress in that particular area.

Given the urgent need for novel materials, we should respect the importance of breakthroughs in materials science for solving thermal problems. The industry still has a long way to go to find materials that are flexible, reliable and economical.

Many different TIMs are being explored that are not based on polymers. For example, the result of sintering silver is a very hard, high thermal conductivity silver alloy matrix between the lid and the mold. Another example is softer metallic materials. When the reliability and other advantages are not there, the phase-change materials that were often talked about a few years ago seem to have disappeared.

To dissipate power in flip-chip packages, sintered silver epoxies exhibit better thermal performance, so people use pressureless (such as Atrox) or pressure sintered epoxies (Argomax). In the flip-chip approach, the heat sink is a nickel-plated copper design that contacts the backside of the chip with a thermal interface material (TIM) at the interface. Other innovations use multiple wires on the back of the chip, which are then connected to the ground plane of the PCB to improve heat dissipation. Copper is still the best thermal interface and is very cost-effective.

This is one of the motivations for working in microfluidics if there is a way that will completely eliminate the need for a TIM. Alternative cooling solutions can avoid the use of interface materials. Moving closer to the chip eliminates these materials. Either improve the materials, or throw them away.

The result of these challenges is that addressing heat issues is increasingly rising up the list of budgetary priorities. Clients are often amazed that they have to spend so much of their budget on thermal energy. Everything comes with a lot of complexity. Treat something new and usually not adopt it until it is proven effective and all liability issues are resolved. Many customers are becoming aware of this and starting to produce products with more advanced engineering, skill and experience.