Immersion Cooling in Silicon Photonics

Immersion Cooling for Data Centers

The need for ever-increasing compute performance for Data Centers is driving innovation across the board in next-generation processors, GPUs, and all other key elements in a network. But these compute performance gains come with dramatic increases in power consumption, posing urgent challenges in terms of power efficiency and heat management.

The graph below shows how power per rack has increased by a factor of 6x in the past 12 years, and accelerating over time.

Immersion Cooling in Silicon Photonics

(image source: Promersion

This is driven by CPUs and GPU that have ever increasingly power dissipation needs with each new generation of chip:
Immersion Cooling in Silicon Photonics

(image source: Promersion

Increasingly, people are looking at cooling options inside Data Centers that a few years ago seemed extreme, including Immersion Cooling. In this article, we’re taking a look at the benefits immersion cooling offers, as well as the obstacles in the way of implementation.
Types of Immersion Cooling: Single-Phase and Two-Phase

Immersion cooling involves submerging computer components in a thermally conductive, but electrically insulating liquid. This method stands out for its exceptional capability to transfer and pull heat away from key electronic components including GPUs, CPUs, and networking components such as switch chips and optical transceivers. Other methods like air cooling, cold plate and passive/active door heat exchangers will continue to exist for certain applications, however immersion cooling will increasingly bridge the gap, offering more efficient ways to preserve hardware for high-compute applications.

This can be done in one of two ways: single-phase immersion cooling or two-phase immersion cooling. Each type has distinct mechanisms and advantages.

Single-Phase Immersion Cooling
Electronic components are submerged in a thermally conductive but electrically insulating liquid. This liquid does not change its state, and remains liquid throughout the cooling process.
How Single-Phase Immersion Cooling Works

Heat Absorption: The cooling liquid absorbs heat directly from the components, like CPUs and GPUs.

Heat Dissipation: The heated liquid is then circulated (usually via a pump) through a heat exchanger, where it releases the absorbed heat.

Recirculation: The cooled liquid is then recirculated back into the system.

This method offers a number of advantages, beginning with consistency: the liquid remains in the same state, making the system relatively simple and consistent in operation. This also means single-phase immersion cooling systems generally require less maintenance than two-phase systems, as there is no risk of evaporation or losing fluid if the tank is opened or not properly sealed. Many of these fluids have a boiling point much higher than water (typically above 200 deg. C.)

Immersion Cooling in Silicon Photonics

(image source: GR Cooling

In addition to simplicity, this method tends to be more cost-effective to set up and operate in comparison to two-phase systems, which we’ll examine next.
Two-Phase Immersion Cooling

Two-phase immersion cooling takes a more dynamic approach. In this system, the cooling liquid undergoes a phase change – it transitions from liquid to gas and back to liquid.

How Two-Phase Immersion Cooling Works

Boiling: When electronic components heat up, they cause the liquid in direct contact with them to boil and vaporize. The boiling point temperature of the liquids used in this approach is typically about 55 deg. C, meaning that the electronic components remain at a stable temperature very close to this level.

Vapor Release: The vapor then rises and comes in contact with a condenser.

Condensation: The condenser cools the vapor, which condenses back into a liquid.

Recycling the Liquid: The condensed liquid is then returned to the main body of liquid to absorb more heat, repeating the cycle.

This method enables highly efficient cooling, because the phase change from liquid to gas absorbs a significant amount of heat. A two-phase system can also provide precise cooling at hotspots, as the liquid directly boils off areas with higher heat loads.

Immersion Cooling in Silicon Photonics

(image source: GR Cooling

Immersion Cooling Methods: Comparison and Use Cases

Two-phase systems are typically more efficient at heat removal due to the phase change process. However, single-phase systems are more efficient in terms of overall energy consumption due to their simpler setup.

Single-phase cooling is often used for less heat-intensive applications, whereas two-phase cooling is preferred for high-density, high-heat generating environments like advanced CPUs and GPUs in data centers.

However, both Microsoft and Meta have halted their research into two-phase immersion cooling, citing health risks. At least for now, it seems that this method may have hit a roadblock.**

Design Challenges: Using Optical Components in Immersion Cooling

Free Space Gaps and Refraction

Many optical components used in data centers, such as transceivers, incorporate free-space gaps in their designs. These gaps enable effective coupling from the laser to the Photonic Integrated Circuit (PIC) and from the PIC to the optical fiber and are inherent to the way many optical components have been designed for the past several decades. But these free-space gaps also pose a number of challenges to immersion cooling systems.

Because liquids (including immersion liquids) have a different index of refraction than air, when a traditional optical component is immersed in fluid, the light path is significantly altered, rendering these devices effectively unusable in that application. This means that either: (1) optical component designs need to be altered to work in an immersion cooling environment as a purposed built- application or (2) the optical components need to be designed to eliminate the free-space gaps such that they can be used both in traditional air cooled applications as well as immersion cooled applications. The problem with first approach (1) is that it requires a second, alternate design for the optical components for the immersion cooling market, increasing the overall design costs and reducing the economies of scale associated with those optical components.

Material Incompatibility

Some materials are simply not suitable for immersion cooling, because they break down in the liquid environment. This is problematic for all liquid cooling systems, since pumps and other recirculating systems are typically required where residue from materials that break-down can get caught.

DustPhotonics’ Material System: Simplicity and Compatibility, Without Free-Space Gaps

At DustPhotonics we have taken steps to sidestep these issues and ensure that our devices can safely undergo immersion cooling. Our products, including our transmit Carmel-8-IMC chip which we showcased at this year’s ECOC event, supports an optical assembly with no free-space optics (see picture below of DustPhotonics live immersion cooling demonstration at ECOC). Inside our chip, lasers are butt-coupled to the PIC and the fiber is also attached directly to the PIC. Thus our products enable transceiver designers to elegantly design a single transceiver that can be used for both immersion and air cooling, and leverage economies of scale in manufacturing.

Immersion Cooling in Silicon Photonics

In addition, our material system uses a straightforward combination of a laser, fiber, and silicon chip, reducing the risk of breakdown in the liquid environment. This simple transmit architecture is less vulnerable to the risks we outlined above.

Contact us to learn more about the DustPhotonics immersion cooling products and how they can be used to support next generation data centers.


Ready to start your Silicon Photonics journey?

© All rights reversed to DustPhotonics Ltd. Terms of Use


Ronnen Lovinger

CEO & Board Member

Python Automation Engineer­

Modiin, Israel