Technical article · Thermal engineering

Two-phase cooling for electronics: performance, limits and test requirements

The transition from a single-phase water circuit to flow boiling changes the heat flux density that a system can achieve. It also changes the requirements for the test bench: flow control becomes a question of stability.

Context

Content and validity

Research statusAugust 2026
Time to readapproximately 25 min
ReadersDevelopment of cold plates and coolant distribution units (CDU), test field
Sources32 documented references

For a test bench for liquid-cooled electronics, there are usually 2 numbers available. These numbers are the power loss of an accelerator and the connected load of a rack. Both numbers are large. Both numbers give little information about the necessary cooling method. The important quantity is seldom in the customer specification: the heat flux density at the contact surface. It decides if a single-phase water circuit is sufficient or if the coolant must boil in the channel.

This article gives the current status: documented performance values, applicable guidelines, physical limits and the measurement tasks for the test bench that follow from them. Profectronics builds test benches and does not operate data centres. Thus the article always describes the view of the test field and not the view of the operator.

Power, power density and installed base

NVIDIA specifies a power input of approximately 120 kW for a GB200 NVL72 rack. 8 power shelves with 33 kW each supply this power^[1]. For the next generation GB300 NVL72, the reference architecture gives “up to 142 kW” per rack^[2]. A system manufacturer gives an operating range of 132 to 140 kW. This manufacturer also supplies manifolds in the rack for up to 250 kW^[3].

The installed base is much lower. For 2025, the Uptime Institute gives an average rack density of slightly less than 9 kW. In the previous year, the value was 8.3 kW. Without outliers, the value is 7.5 kW. More than 80 % of the operators in the survey state that their facility has no rack above 30 kW^[4].

Between new facilities and the installed base, there is a factor of approximately 15. Thus a test bench for both ranges must be adjustable over this factor and not operate at only one operating point.

The transition is also visible on the air side. A recent technical publication gives approximately 25 to 30 kW per rack for conventional air systems. New loads are frequently above 100 kW^[7]. In 2021, ASHRAE (American Society of Heating, Refrigerating and Air-Conditioning Engineers) identified a limit of water cooling with cold plates at approximately 500 W per processor. At that time, ASHRAE gave a rack with 50 kW as the high-density case^[5].

Heat flux density instead of rack power

Rack power is not a suitable quantity for the design of a cooling method. The important quantity is the heat flux density in W/cm² at the surface through which the heat flows out. At this point, the documented data ends. NVIDIA does not publish die areas. Without the area, each conversion of the thermal design power (TDP) into W/cm² is an estimate. Thus this article gives only heat flux densities that a publication measured on a defined heater surface.

Numbers in this article. Each number in this article has a source. If the research did not find the primary source of a frequently quoted number, the number is not in this article. This applies mainly to comparison tables of peak heat flux densities. The literature passes these tables on without measurement conditions that a reader can verify.

Guidelines of ASHRAE, OCP and JEDEC

ASHRAE: water classes

The applicable classification of supply temperatures is in ASHRAE Datacom Book 1^[6]. The white paper of TC 9.9 that accompanies it gives the change of class names from 2021 word for word^[5]:

W1 became W17.
W2 became W27.
W3 became W32.
W40 is an additional class.
W4 became W45.
W5 became W+.

The classes apply to the supply temperature of the facility water system (FWS)^{[5], [7]}. The number in the class name is the maximum permitted supply temperature in °C. For all classes, the lower limit is 2 °C^[5]. W+ is above the class W45. Downstream of the CDU, the supply temperature to the IT is higher by the temperature difference of the heat exchanger.

Water classes to ASHRAE TC 9.9, supply of the facility water system
Class	Supply temperature	Result for the test bench
W17	2 to 17 °C	The test bench needs its own chiller. The test bench must monitor the dew point in the test room as a safety function.
W27	2 to 27 °C	The usual design point for cold plate tests today.
W32	2 to 32 °C	Free cooling during the full year is realistic in Central Europe.
W40 / W45	2 to 40 / 45 °C	The task is control quality, not cooling capacity. The test bench must heat as well as cool.
W+	above 45 °C	Specific to the application. A design is possible only with characteristic values of the test specimen.

Test bench designs most frequently miss the requirement of the classes W40 and W45. A test bench for W40 needs a heater with sufficient capacity in the supply line. On a cold winter day, it must actively bring the fluid to temperature. A design only for cooling capacity does not include this operating case.

Open Compute Project: test specifications

For the test itself, the Open Compute Project (OCP) gives more information than ASHRAE because it specifies acceptance criteria. The 2019 requirements document for cold plates refers to 2 regimes for the pressure test. IEC 62368-1 requires leak tests at 3 times the normal operating pressure and at 2 times the pressure in the fault case. ASME B31.3 requires them at 1.5 times the design pressure. The typical operating pressure range is 140 to 450 kPa. To prevent erosion, the specification gives a flow velocity in the pipe of 1.5 to 2.1 m/s^[8].

The white paper on qualification is more specific^[9]. For test bench builders, it is the most useful OCP document that is publicly available. It describes tests with acceptance criteria:

Tests and acceptance criteria to OCP Cold Plate Development and Qualification
Test	Requirement	Reference
Leak tightness	Pressure drop less than 0.5 %, with an additional bubble test	EN 1779
Proof pressure	1× maximum operating pressure, 5 min hydrostatic	not specified
Burst pressure	3× maximum operating pressure, 2 min	not specified
Thermal resistance	R = (T_c − T_L) / Q, with a defined thermal interface material	ASTM D5470^[14]
Flow velocity	in the cold plate less than 1.5 m/s	not specified
Galvanic corrosion	Potential difference metal/metal maximum 0.15 V	not specified
Salt spray	35 °C, 5 % NaCl, pH 6.5 to 7.2, 8 h, no corrosion permitted	ASTM B117
Flatness and roughness	of the contact surface to the component	ISO 12781-2 · ISO 21920-2
Fluid condition	Metal ions, pH, reserve alkalinity, chloride	ASTM D6130 · D5185 · D1287 · D1121 · D5827

There are 2 observations:

OCP uses only a relative pressure drop criterion and the bubble test. The documents that are publicly available do not specify a quantitative leak rate in mbar·l/s, for example from a helium test. A person who writes such a number into a customer specification must give the justification for it.
Fluid analysis to ASTM has the same importance as the thermal tests. The condition of the coolant over the service life is a test item and not an operation detail.

JEDEC: thermal measurement

For the thermal resistance between junction and case, JESD51-14 is the applicable standard. The standard describes the transient measurement with 2 interfaces. With the structure function, it is possible to isolate the conductive part from this measurement^[11]. JESD51-1 specifies the electrical measurement of the junction temperature and the calibration of the temperature-sensitive parameter^[12]. JESD51-4A applies to the chip emulator^[13].

No standard for flow boiling. The JESD51 series covers natural and forced convection and the cold plate. The research found no JEDEC document that standardises flow boiling as a measurement environment. Thus there is no standardised boundary condition for transient thermal resistances on a cold plate with flow boiling. Then the boundary condition is part of the test specification and must be in the test report.

Physics of flow boiling

Two-phase cooling uses the enthalpy of vaporisation. While the fluid boils, it absorbs heat at an approximately constant temperature. The wall temperature follows the saturation temperature and not the integrated temperature rise of the single-phase circuit. Thus a two-phase circuit gives more uniform case temperatures across a row of cold plates than a water circuit. In the water circuit, the last test specimen gets the fluid after the first test specimen heated it.

Onset of nucleate boiling

The transition does not start suddenly. The point at which the first vapour bubble can grow on the wall depends on wall superheat, nucleation site size and contact angle. An open-access paper gives measured data from microchannels with a width of 275 µm and a depth of 636 µm. For water, it gives measured heat flux densities at the onset of boiling of 9.88 to 68.81 W/cm² over 23 test cases. The related model agrees with the measurement within ± 20 %^[15].

For the test bench, this is an advantage: the onset of boiling is far below the target operating points and is itself a measurable quantity. Thus it is a test point and not an uncertainty. A fast temperature measurement is necessary. It must resolve the typical peak of the wall superheat when boiling starts.

Critical heat flux

The critical heat flux (CHF) sets the upper limit. Above this limit, a continuous vapour film forms on the wall. The heat transfer collapses and the wall temperature jumps. For pool boiling, Zuber gives a hydrodynamic explanation of this limit^[21]. For forced flow in tubes, the generalised correlation of Katto and Ohno describes it^[20]. For heat transfer in saturated boiling, the methods of Chen^[19] and Kandlikar^[18] are the usual references.

In parallel microchannels, this classical description is not sufficient. There, a flow instability usually sets the limit and not the hydrodynamic limit. The instability starves a single channel of flow long before the mean heat flux density becomes critical. Section 04 describes this process.

Ledinegg instability in parallel channels

The pressure drop characteristic of a heated channel is not monotonic. It has the shape of an N with 3 branches:

At low mass flow, most of the fluid evaporates. Mostly vapour flows in the channel, and the pressure drop increases with the mass flow.
When the mass flow increases, the vapour quality and the specific volume decrease. The pressure drop decreases.
At high mass flow, the friction of the liquid is dominant, and the characteristic increases again.

Between the maximum and the minimum, there is a range with a negative slope.

Each manifold applies a common pressure difference to a group of parallel channels. This pressure difference can intersect the characteristic at 3 points: on the vapour branch, the middle branch and the liquid branch. The middle intersection point is unstable. A small disturbance makes the operating point jump suddenly to the hot branch. The name of this static instability comes from Ledinegg. In the classification, it is adjacent to the dynamic types, for example density wave oscillation and pressure drop oscillation^[17].

Fig. 1 · Ledinegg excursion, schematic. With an imposed pressure difference, the characteristic of the heated channel has 3 intersection points. Point B is unstable. The operating point jumps to A: low flow, much vapour and a higher wall temperature. An inlet restrictor makes the characteristic so steep that only 1 intersection point remains. Schematic diagram, not a measured curve.

A measurement on thermally isolated parallel microchannels shows the size of the effect. The power increased by 0.6 W, from 2.6 to 3.2 W. During this increase, the wall temperature of the starved channel jumped from 124 °C to 219 °C. The mean mass flux decreased only from 400 to 368 kg/(m²·s). The pressure drop increased from approximately 3.5 to 11.7 kPa^[16].

A load change of 0.6 W causes a temperature jump of 95 K. Thus a two-phase test bench must measure the temperature quickly and in each branch. A mean value across the row of cold plates does not show this process.

Inlet restrictor as countermeasure

The usual method is the inlet restrictor: an orifice or capillary upstream of each channel. Its single-phase pressure drop increases with the square of the flow. It raises the part of the characteristic with negative slope until the total characteristic increases monotonically. Then only 1 intersection point remains^{[17], [16]}. The restrictor needs additional pump power. For the test bench, it adds one more unknown: its pressure drop occurs between the measurement point and the cold plate.

This gives a design rule for the test bench: the pressure measurement point for the cold plate must be downstream of the restrictor. If the test bench records the manifold pressure as the inlet pressure of the cold plate, the measurement includes the restrictor loss. The calculated thermal resistance then changes at the next orifice change, although the cold plate stays the same.

Comparison of design types

The overview contains only values that the given source measured on a defined heater surface. Thus it is not complete. For spray cooling and conventional vapour chambers, the applicable papers are behind publisher paywalls. The article gives only numbers with a measurement condition that the reader can trace.

Documented characteristic values per design type
Design type	Medium	Heat flux density	Thermal resistance	Boundary condition
Microchannel, single-phase	Water	200 to 600 W/cm²	0.23 to 0.71 K/W	Heater surface 10 to 30 mm², pressure drop 17 to 400 kPa^[24]
Microchannel with jet impingement, single-phase	Water	400 W/cm²	0.11 K/W	Heater surface 210 mm², 1.90 L/min, Δp 13.5 kPa^[24]
Cold plate, two-phase	R1233zd(E)	250 W/cm²	0.017 K/W at 1 kW	4 heaters with 1 cm² each, outlet vapour quality ≈ 0.55^[22]
Rack circuit, two-phase	R-134a	100 W/cm²	≤ 0.012 K/W	Rack with 50 kW, vapour quality up to 58 %, 0.48 L/min per kW^[23]
Capillary hybrid evaporator	R245fa	CHF 970 W/cm²	0.09 K·cm²/W	above 850 W at less than 1.0 W pump power^[25]

The table shows the advantage of two-phase operation. This advantage is not the peak heat flux density. The single-phase microchannel without jet impingement achieves comparable and higher values. But for these values, it needs pressure drops of up to 400 kPa^[24].

With jet impingement, the single-phase microchannel achieves 400 W/cm² at 13.5 kPa^[24]. The two-phase cold plate achieves 250 W/cm² at 0.017 K/W^[22]. The two-phase rack circuit has a pressure drop of approximately 76 kPa (11 psi) over the full circuit^[23]. Thus the important characteristic value is the removed heat per watt of pump power.

Immersion cooling: field report

Two-phase immersion cooling frequently has the status of the final stage of the development. The field report of a US national laboratory shows a different result. The facility in the study used a fluoroketone with a phase change at 49 °C. The cooling energy decreased by 72 %. This is much more than the target of 50 %^[26].

The facility missed nearly all other targets^[26]:

The processor temperatures were 20 K higher than with conventional cooling.
The compute performance decreased by 5.6 %.
The fluid losses cost 368 % of the IT energy costs.
The payback period was 33 years.
Because of failures, the maintenance work increased many times.

The conclusion of the report: two-phase immersion with this fluid is “not viable at this time”^[26].

The report is from 2016 and is not a current assessment. But it shows where the problems are: in fluid loss, maintainability and cost, not in the thermal performance. Thus, for an immersion test bench, Profectronics recommends to record the fluid balance as a measured quantity and not as a consumable item.

The standards cover immersion comparatively well. The OCP requirements include these items^[10]:

dielectric strength minimum 6 kV/mm
flash point minimum 150 °C
temperature measurement at a minimum of 2 points with ± 1.0 K
containment volume 100 % of the largest system volume
for two-phase systems, a pressure sensor or vacuum sensor is mandatory

Fluids: PFAS restriction and consequences

For all two-phase technology, there is a regulatory question that engineering cannot solve. In December 2022, 3M announced the end of its manufacture of per- and polyfluoroalkyl substances (PFAS) by the end of 2025. At the same time, the company announced the removal of PFAS from its full product portfolio^[27]. The company states that the exit was complete at the end of 2025. Stock from earlier production is still on the market^[28].

Clarification on the press release. The press release does not name Novec or Fluorinert. It refers in general to fluoropolymers, fluorinated fluids and PFAS-based additives^[27]. Thus the primary source does not support the frequent statement “3M stops Novec”. This statement comes from the communication of distributors.

In parallel, the European restriction procedure continues. 2 independent legal analyses give the same sequence^{[29], [30]}:

2026-03-02: The Committee for Risk Assessment adopted its final opinion.
2026-03-26: Draft of the socio-economic opinion.
Until 2026-05-25: Consultation.
End of 2026: Final opinion expected.
2027: Adoption of an amendment to REACH Annex XVII expected, with a transition period of 18 months.

For heat transfer fluids, the procedure explicitly leaves the status open. The Committee for Socio-economic Analysis considers the data insufficient. The data does not permit a final assessment of the requested derogations for refrigerants, heat transfer fluids and electronic applications^[29]. For high-technology sectors without available alternatives, time-limited derogations of up to 12 years are under discussion^[30].

Consequences for the fluid selection

The obvious substitutes do not solve the problem. A development product that replaces perfluoropolyethers, hydrofluoroethers and fluoroketones boils at 49 °C. It has a global warming potential of 10 and no flash point. Chemically, it is a hydrofluoroolefin^[31]. Thus the product itself is a fluorinated compound. Under the broad definition of the European proposal, it is in the scope of the proposal.

The same applies to R1233zd(E). The test set-up in section 05 uses this fluid^[22].

This research did not find a PFAS-free two-phase fluid with comparable properties for electronics cooling.

The consequence for plant engineering: the fluid selection must stay reversible. In practice, this means:

The saturation pressure range of the circuit permits more than one candidate.
Materials and seals must have test results for all fluids on the list of candidates and not only for one fluid.
The fill quantity is such that a fluid change stays a maintenance task and does not need a conversion.

Requirements for the test bench

Fig. 2 · Pumped two-phase circuit for direct-to-chip tests, schematic to the symbols of ISO 10628. The inlet restrictors upstream of the parallel cold plates decide the stability of the full test line.

Measured quantities

The test specifications and publications in this article give a set of measured quantities for a two-phase test bench. This set is longer than for a single-phase circuit because the condition of the fluid itself becomes a measured quantity.

Measured quantities and their basis
Quantity	Purpose	Basis
Heat flux density	Reference quantity for each statement, meaningful only with a defined heater surface	JESD51-4A
Thermal resistance case to fluid	Characteristic value of the cold plate	OCP Qualification
Transient thermal resistance	Separation of the heat paths with the structure function	JESD51-14
Junction temperature	electrical, with a calibrated temperature-sensitive parameter	JESD51-1
Pressure drop per branch	downstream of the inlet restrictor, not upstream	Section 04
Mass flow per branch	Find maldistribution before it becomes thermally visible	OCP Qualification
Saturation pressure and saturation temperature	set the operating point of the full circuit	OCP Immersion Rev. 2.10
Outlet vapour quality	from the energy balance, limits the permitted load	Energy balance
Leak tightness	Pressure drop less than 0.5 %, bubble test	EN 1779 · OCP
Fluid condition	Metal ions, pH, chloride over the service life	ASTM D6130 · D5185 · D5827
Contact resistance of the thermal interface material	Comparability between test runs	ASTM D5470

Chip emulator

A real processor is not suitable as a test specimen. Its power input depends on the load, and its heat distribution is unknown. Also, it is not possible to set it to a fixed operating point. Thus a test of cold plates with valid results starts with a thermal test vehicle (TTV). JESD51-4A gives 3 purposes for it^[13]:

a defined structure that generates heat flux and has integrated temperature sensors
the validation of thermal simulation models
the study of complex assemblies

The design requirements of the standard apply directly to test bench construction^[13]:

Metal film resistors heat more uniformly, with less than 1 % temperature variation over 100 °C. Thus the standard prefers them.
Transistors permit a tighter power control, less than 0.5 %. But in large cells, they heat non-uniformly.
For cells with an edge length of 2.5 mm or more, the standard recommends a heater surface of a minimum of 85 % of the cell area. For smaller cells, the recommendation is a minimum of 60 %.
The temperature sensor is a forward-biased diode with a calibration factor of approximately −0.5 K/mV at approximately 1 mA measurement current.
The standard recommends a diode power loss of less than 1 % of the heater power. Thus the diode does not heat itself.
The geometric tolerance relative to the simulated component is ± 10 % in thickness and area.

A real test set-up used 4 ceramic heaters with 250 W each on 1 cm² each, below a copper heat spreader. Thermocouples were directly above the heaters. The set-up simulated a generation of server processors^[22].

Measurement uncertainty and its sources

2 published uncertainty budgets from two-phase test benches give the realistic range. The first set-up gives^[22]:

± 0.5 K for type T thermocouples
± 1 % for the absolute pressure
0.08 % for the differential pressure
0.3 % repeatability for the volume flow
less than 10 W at 1 000 W heater power

At 50 kW rack load, the second set-up gives ± 0.0002 K/W for the thermal resistance. At a measured value of 0.012 K/W, this is approximately ± 1.7 %. For the outlet vapour quality, it gives ± 1.83 %^[23]. The propagation of uncertainty usually follows the Guide to the Expression of Uncertainty in Measurement (GUM)^[32].

Assessment by Profectronics, not a statement of the sources. The thermocouples have an uncertainty of ± 0.5 K. Typical temperature differences between case and fluid are 15 to 20 K. Thus the temperature measurement is the largest part of the uncertainty budget for the thermal resistance. Type T thermocouples are not sufficient to resolve the thermal resistance to ± 2 %. For this resolution, calibrated Pt1000 elements or an in-situ calibration at the zero-power point are necessary.

Conclusions for the test bench design

Sections 01 to 07 give 7 points for a test bench for cold plates and manifolds of liquid-cooled electronics. These points are conclusions of Profectronics and not of the sources in this article.

The chip emulator is a core component. Without a defined heater surface and a calibrated temperature measurement in the emulator, a value in W/cm² has no significance. JESD51-4A gives the design limits.
Measure each branch, not the total. A mean value across 4 parallel cold plates hides the process that puts the system at risk. The Ledinegg excursion shows first in a single branch.
Measure quickly. A sample interval of 1 s cannot resolve a temperature jump of 95 K at a load change of 0.6 W. The data acquisition must operate in the range of tenths of a second.
Inlet restrictor in the functional specification. The restrictor must be replaceable. Its pressure drop must have its own measurement. If not, this pressure drop becomes part of the thermal resistance of the test specimen.
Saturation pressure as the controlled variable. The saturation pressure sets the saturation temperature and thus the full thermal operating point. A two-phase circuit without pressure control has no operating point.
Keep the fluid selection reversible. While the European restriction procedure continues, each decision for one fluid is a risk. Materials, seals and pressure range follow a list of candidates.
Leak tightness as a routine test. The OCP criterion (less than 0.5 % pressure drop and bubble test) is a test of each component, not a random sample. For a circuit that moves a dielectric fluid across electronics, this test level is appropriate.

This article does not recommend for or against two-phase cooling. The decision depends on 3 quantities. These are the heat flux density of the specific component, the available heat rejection and the legally secure availability of the fluid. A test bench for both methods stays usable after each of these decisions. In the assessment of Profectronics, the additional cost occurs mainly in the measurement chain.

Validity: All data refer to the research status of August 2026. The procedure status of the European PFAS restriction changes. The dates in this article come from 2 independent legal analyses. Before an investment decision, compare these dates with the publications of the European Chemicals Agency (ECHA).

Profectronics builds test benches and does not certify. An accredited body makes conformity statements about test specimens.

Sources

All references retrieved in August 2026. If a source was only available through a mirror or had limited full-text access, the article does not use its numbers.

NVIDIA · NVIDIA DGX GB Rack Scale Systems User Guide — Hardware · version 2026 · docs.nvidia.com/dgx/dgxgb200-user-guide/hardware.html · manufacturer documentation
NVIDIA · Enterprise Reference Architectures — NVL72 AI Factory, System Hardware & Components · retrieved 2026 · docs.nvidia.com/enterprise-reference-architectures/nvl72-ai-factory/latest/components.html · manufacturer documentation
Supermicro · NVIDIA GB300 NVL72 · SuperCluster Datasheet · 2025 · www.supermicro.com/datasheet/datasheet_SuperCluster_GB300_NVL72.pdf · manufacturer documentation
Uptime Institute · Global Data Center Survey 2025 · July 2025 · datacenter.uptimeinstitute.com/rs/711-RIA-145/images/2025.Annual.Survey.Report.pdf · market study
ASHRAE TC 9.9 · Emergence and Expansion of Liquid Cooling in Mainstream Data Centers (White Paper) · 2021 · www.ashrae.org/file%20library/technical%20resources/bookstore/emergence-and-expansion-of-liquid-cooling-in-mainstream-data-centers_wp.pdf · technical committee
ASHRAE · Datacom Book 1: Thermal Guidelines for Data Processing Environments, 5th Edition · 2021 · ISBN 9781955516792 · webstore.ansi.org/standards/ashrae/ashraedatacombook5thedition · technical guideline
CIBSE Journal · Andy Pearson · Module 254: Liquid cooling in data centre applications · October 2025 · www.cibsejournal.com/cpd/modules/2025-09-lcdca/ · technical journal
Open Compute Project · ACS Liquid Cooling Cold Plate Requirements Document, Revision 1.0 · 2019-10-09 · www.opencompute.org/documents/ocp-acs-liquid-cooling-cold-plate-requirements-pdf · consortium specification
Open Compute Project · White Paper: Cold Plate Development and Qualification · no date in the document header · www.opencompute.org/documents/ocp-cold-plate-development-and-qualification-with-integrated-comments-pdf · consortium specification
Open Compute Project · ACS Immersion Requirements, Rev. 2.0 / amended 2.10 · 2023-08-18 · www.opencompute.org/documents/ocp-acs-immersion-requirements-rev-2-1-pdf · consortium specification
JEDEC · JESD51-14 · Transient Dual Interface Test Method for the Measurement of the Thermal Resistance Junction-to-Case · November 2010 · www.jedec.org/standards-documents/docs/jesd51-14-0 · standard
JEDEC · JESD51-1 · Integrated Circuit Thermal Measurement Method — Electrical Test Method · December 1995 · www.jedec.org/standards-documents/docs/jesd-51-1 · standard
JEDEC · JESD51-4A · Thermal Test Chip Guideline (Wire Bond and Flip Chip) · June 2019 · www.thermengr.net/Stds/JESD51-4A.pdf · standard
ASTM International · ASTM D5470-17(2024) · Standard Test Method for Thermal Transmission Properties of Thermally Conductive Electrical Insulation Materials · 2024 · store.astm.org/d5470-17.html · standard
D. Liu, P.-S. Lee, S. V. Garimella · Prediction of the onset of nucleate boiling in microchannel flow · Int. J. Heat Mass Transfer 48 (2005) 5134–5149 · www2.egr.uh.edu/~dli9/2005_ijhmt_onb.pdf · peer-reviewed
T. A. Kingston, J. A. Weibel, S. V. Garimella · Ledinegg Instability-Induced Temperature Excursion between Thermally Isolated, Heated Parallel Microchannels · Int. J. Heat Mass Transfer, 2019 · docs.lib.purdue.edu/cgi/viewcontent.cgi?article=1333&context=coolingpubs · peer-reviewed
Thermopedia (Begell House) · Two-Phase Instabilities · DOI 10.1615/AtoZ.t.two-phase_instabilities · www.thermopedia.com/content/25/ · technical encyclopaedia
S. G. Kandlikar · A General Correlation for Saturated Two-Phase Flow Boiling Heat Transfer Inside Horizontal and Vertical Tubes · J. Heat Transfer 112 (1) 1990, p. 219 ff. · www.osti.gov/biblio/5633093 · peer-reviewed
J. C. Chen · Correlation for Boiling Heat Transfer to Saturated Fluids in Convective Flow · I&EC Process Design and Development, 1966 · pubs.acs.org/doi/10.1021/i260019a023 · peer-reviewed
Y. Katto, H. Ohno · An improved version of the generalized correlation of critical heat flux for forced convective boiling in uniformly heated vertical tubes · Int. J. Heat Mass Transfer 27 (1984) · ui.adsabs.harvard.edu/abs/1984IJHMT..27.1641K/abstract · peer-reviewed
N. Zuber · Hydrodynamic Aspects of Boiling Heat Transfer · Report AECU-4439, OSTI-ID 4175511 · www.osti.gov/biblio/4175511 · research report
Q. Wang, S. Ozguc, A. Narayanan, R. W. Bonner III · A Server-Level Test System for Direct-To-Chip Two-Phase Cooling of Data Centers Using a Low Global Warming Potential Fluid · 23rd IEEE ITHERM Conference, 2024 · accelsius.com/wp-content/uploads/A-Server-Level-Test-System-A2-03-1-049.pdf · conference paper
A. Heydari et al. · Advancing in Data Centers Thermal Management: Experimental Assessment of Two-Phase Liquid Cooling Technology · ASME InterPACK2024, DOI 10.1115/IPACK2024-141342 · par.nsf.gov/servlets/purl/10611032 · conference paper
R. Xiao et al. · Experimental Study on Cooling Performance of a Hybrid Microchannel and Jet Impingement Heat Sink · Applied Sciences 12 (24) 13033, 2022 · www.mdpi.com/2076-3417/12/24/13033 · peer-reviewed
M. R. Shaeri, R. W. Bonner, M. C. Ellis · Thin Hybrid Capillary Two-Phase Cooling System · Advanced Cooling Technologies, 2020 · www.osti.gov/servlets/purl/1803455 · research report
H. Coles, M. Herrlin · Immersion Cooling of Electronics in DoD Installations, LBNL-1005666 · Lawrence Berkeley National Laboratory, May 2016 · datacenters.lbl.gov/sites/default/files/ImmersionCooling2016.pdf · research report
3M · 3M to Exit PFAS Manufacturing by the End of 2025 (press release) · 2022-12-20 · news.3m.com/2022-12-20-3M-to-Exit-PFAS-Manufacturing-by-the-End-of-2025 · company statement
3M · PFAS Stewardship — Uses and Applications · retrieved 2026 · www.3m.com/3M/en_US/pfas-stewardship/uses-applications/ · company statement
White & Case LLP · Europe's PFAS restriction proposal is moving forward · 2026 · www.whitecase.com/insight-alert/europes-pfas-restriction-proposal-moving-forward · legal analysis
Arnold & Porter · ECHA Committees Advance Broad PFAS Restriction Under REACH · March 2026 · www.arnoldporter.com/en/perspectives/advisories/2026/03/echa-committees-advance-broad-pfas-restriction-under-reach · legal analysis
The Chemours Company · Opteon™ 2P50 Developmental Dielectric Heat Transfer Fluid · retrieved 2026 · www.opteon.com/en/products/liquid-cooling/2p50 · manufacturer documentation
BIPM / JCGM · JCGM 100:2008 — Guide to the expression of uncertainty in measurement (GUM) · 2008 ff. · www.bipm.org/en/committees/jc/jcgm/publications · technical guideline

Enquiry

Test of cold plates and manifolds

Give the heat flux density, the fluid, the saturation range and the operating mode: single-phase, two-phase or both. We clarify the other points in the functional specification.

Send an enquiry Focus area data centre cooling All technical articles