A Novel Functional Accelerated Thermal Aging Methodology for Single-Phase Dielectric Immersion Coolants

Abstract

High-performance computing is significantly transforming the thermal management landscape in data centers. Immersion cooling presents a promising solution to efficiently cool high-density server racks while drastically lowering the energy footprint of this industry. Coolant research is key in the global deployment of immersion cooling systems, although these studies remain rather scarce. From an operational perspective, the coolant quality over time and corresponding lifespan are key concerns. However, little is known about how these parameters behave under actual server operating conditions. Therefore, this work introduces a comprehensive aging methodology to evaluate the quality of dielectric immersion coolants over time and predict their lifespan in single-phase immersion cooling applications. The approach involves the accelerated thermal aging of a dielectric immersion coolant at 100 °C for six months. Subsequently, the thermo-electrical performance of the aged coolant is validated using an immersed Raspberry Pi 4 Computer. A biobased synthetic ester dielectric coolant (L759, Oleon NV) was selected because such chemistry can effectively balance crucial performance and safety metrics. Its low viscosity supports efficient heat transfer, as demonstrated by the high Mouromtseff number and thermal figures of merit 1–3, all of which exceeded established threshold values. Also, L759 exhibited a high flash point, high electrical resistivity, and high dielectric breakdown, conforming to fire and electrical safety guidelines. Furthermore, the coolant showed minimal interactions with polyvinyl chloride, a widely used cable insulation material. Additionally, the product is safe to handle and has an excellent environmental profile. Due to the careful selection of use case materials, the proposed system-oriented aging methodology successfully simulated actual material–liquid interactions. Thermal aging and use case data further identified the acid value (physical; primary), specific heat capacity (thermal; secondary), and electrical resistivity (electrical; secondary) as crucial coolant lifespan indicators. Furthermore, the aged coolant properly cooled the central processing unit on the Raspberry Pi 4 Computer, as reflected in thermo-physical and dielectric/electrical property data. No material incompatibilities were observed during functional testing. Overall, L759 can comply with the lifespan of current (five years) and next-generation (seven, up to 10 years) servers. Hence, ester technology shows great potential in operating single-phase immersion cooling systems. The presented methodology integrated material compatibility, thermal stress, and functional performance factors. It can easily be extrapolated to alternative coolant chemistries and even to adjacent application domains, thereby contributing to standardization efforts across the broader industry.

1. Introduction

High-performance computing is rapidly emerging as a critical technology across a wide range of applications in both academia and industry. Within this field, advancements in artificial intelligence, machine learning, and quantum computing are driving the demand for high-density server racks, with power requirements up to 100 kW and more. Moreover, as digitalization continues to grow, the digital industry is projected to drive more than 20% of the growth in the world’s energy demand by 2030. , Recently, in alignment with UN’s Sustainable Development Goals, regulatory bodies worldwide are pushing this sector to drastically lower its energy footprint. , To meet these sustainability goals, thermal management in data centers has become a major field of study. −

Efficient cooling can significantly reduce the total energy consumption of a data center while guaranteeing high-performance and long-term system reliability. Despite the predominance of air-cooled servers, this cooling technology is facing challenges in terms of cooling capacity (cf. difficulties to cool high-density server racks), infrastructure (e.g., space constraints), and water usage. Among the various cooling technologies, , single-phase (1P) immersion cooling (IC) has emerged as a promising alternative to conventional air cooling. In contrast to two-phase IC, the dielectric coolant does not change state upon functioning. Hence, 1P IC systems are considered less complex, less expensive, and safer to operate. ,

The choice of dielectric coolant is critical for achieving high-performance (1P) IC systems. , The coolant must efficiently cool the server, be nonconductive, and be chemically stable. Fluorochemicals − dominated the coolant market for years, although some of these products will phase out in the following years due to serious human health and sustainability concerns. , Hydrocarbons − and esters are emerging, but scientific coolant research is still scarce. IC studies exclusively reported on flow and related heat transfer mechanisms to optimize the rack (system) design and operating conditions ,− rather than on the coolant itself.

From an operational perspective, the coolant quality over time and related lifespan are key concerns. For (1P) IC systems, the coolant lifespan is aimed at five years (preferably up to 10 years) to align with a typical server lifespan. However, empirical data on coolant longevity under actual server operating conditions are limited.

To the best of our knowledge, only two prior studies , addressed coolant longevity. Tank original equipment manufacturer (OEM) Submer determined the lifespan of a gas-to-liquid (GTL) coolant using the IEC 61125 oxidation test (120 °C, 500 h, including oxygen injection). The coolant was benchmarked against a typical synthetic hydrocarbon IC liquid. Changes in acid value and dielectric dissipation factor were studied and found to be 50% and 10% less pronounced, respectively, compared to the benchmark. The coolant lifespan was not explicitly mentioned, but the authors stipulated that the test simulated 14 years of actual in-service performance. In the second case, server OEM Intel and the Japanese telecommunications operator KDDI demonstrated the feasibility of IC by immersing a 2U modular server chassis in a 12U immersion tank filled with a hydrocarbon coolant. Server performance, coolant specifications, and liquid–material interactions were studied. Moreover, accelerated tests, not described in detail though, were performed to assess the coolant stability. A synthetic polyalphaolefin oil (PAO8) served as the benchmark. The selected coolant exhibited a less pronounced increase in acid value compared to PAO8. Based on the change in acid value solely, the authors claimed a coolant lifespan of 10 years.

Furthermore, in analogous applications such as transformers, − the longevity of insulating liquids is often studied via thermal (accelerated) aging. Gutiérrez et al., for example, elucidated changes in physical and electrical characteristics of four natural esters (sunflower, rapeseed, soybean, and palm oil), a synthetic ester, and a mineral oil upon thermal aging at 150 °C for 4 weeks in the presence of Kraft insulating paper. Changes in liquid characteristics were more pronounced for the screened esters compared to the mineral oil, whereas the opposite trend was observed regarding the degradation of the insulating paper, likely due to the higher moisture tolerance of the ester oils. In addition, the authors highlighted the lack of standardization within this domain.

Batteries in electrical vehicles can also be cooled by immersing them in a dielectric coolant. − Tormos et al. studied the behavior of four dielectric coolants under thermal and electrical stress: mineral GIII, PAO, and two ester (diester, polyol ester) base oils. Thermal aging was performed at 150 °C for 120 h with a Cu strip asa catalyst, while 1000 breakdown voltage discharges were applied to electrically age the dielectric coolants. Among the studied thermo-physical properties, only the kinematic viscosity significantly changed under thermal stress. The impact of electrical stress on the thermo-physical properties was negligible. In contrast, thermal aging strongly affected the electrical properties such as the dielectric dissipation factor and electrical resistivity, whereas the dielectric breakdown voltage was not significantly impacted. For all screened coolants, the dielectric dissipation factor did not change under electrical stress. For hydrocarbons, the electrical resistivity and dielectric breakdown voltage remained unchanged. For esters, the latter trends were less clear, as the water content in these base oils increased upon electrical aging. The electrical resistivity and dielectric breakdown voltage seemed to be slightly decreased, but it was not clear if it was due to the electrical stress itself, the increased water content, or a combination thereof.

To conclude, the literature clearly lacks comprehensive and standardized methodologies to accurately assess the long-term quality and corresponding lifespan of dielectric immersion coolants in 1P IC applications. Under actual operating conditions, the coolant longevity is governed by a complex interplay of factors including the intrinsic liquid characteristics, multiple liquid–material interactions, system-level operating conditions (temperature, frequency, and load cycles) and even environmental factors (humidity). Novel test methods should fully account for their multifactorial nature. Moreover, neither experimental nor simulated performance data for aged coolants are available. In addition, the aging temperature must be carefully selected, as too high temperatures can introduce secondary effects, i.e., effects that would not occur under actual operating conditions.

This work aims to address these critical gaps by presenting a comprehensive, system-oriented aging methodology. Such an approach must offer a rigorous and reliable framework for evaluating the longevity of dielectric immersion coolants in 1P IC applications, thereby minimizing safety risks and the likelihood of system failures during expensive, long-term use case tests. Furthermore, it will contribute to ongoing standardization efforts in the field.

For this study, a biobased synthetic ester dielectric immersion coolant (L759, Oleon NV) was selected because such chemistry can adequately balance a low viscosity for high heat transfer against a high flash point for high fire safety, along with acceptable polyvinyl chloride (PVC) interactions. , Moreover, esters are generally classified as nonhazardous, making them safe to handle. They also exhibit an excellent environmental profile: readily biodegradable, no aquatic toxicity, no global warming potential (GWP), and no ozone depletion potential (ODP).

The coolant was thermally aged at 100 °C for six months in the presence of use case materials to simulate the actual liquid–material interactions. Changes in physical (density, dynamic viscosity, acid value, flash point, water content, and color), thermal (specific heat capacity and thermal conductivity), and electrical properties (electrical resistivity and dielectric breakdown) were monitored over six months. For unused versus aged L759, the corresponding thermal and dielectric/electrical performance metrics were also discussed. Additionally, critical life-cycle indicators were identified. Next, the thermo-electrical performance of the aged coolant was assessed by immersing a single-board Raspberry Pi 4 Computer. The CPU temperature, frequency, and power load were monitored over time for 3 h and benchmarked against both air cooling and immersion cooling with unused dielectric coolant. Furthermore, the data from a recent case study involving precision immersion cooling were used to validate the obtained aging data.

2. Experimental Section

The developed methodology comprises functional accelerated thermal aging experiments, as schematically illustrated in Figure . Functional testing relates to 1P IC. Data of a recent use case were used to validate the experimental results obtained in this work.

1.

Schematic overview of the proposed functional thermal accelerated aging methodology.

The following sections outline

• i.the rationale for selecting the ester-based dielectric immersion coolant;
• ii.the methods used for coolant characterization;
• iii.the accelerated thermal aging protocol; and
• iv.the functional testing procedure.

2.1. Dielectric Immersion Coolant

A biobased synthetic ester dielectric immersion coolant (L759, Oleon NV) was selected because the kinematic viscosity was properly balanced against the flash point and PVC compatibility; see also the Results and Discussion, section . Moreover, ester chemistry has an outstanding safe and environmentally friendly profile.

2.1.1. Coolant Characteristics

Physical, thermal, and electrical properties are crucial in assessing the performance of a particular dielectric immersion coolant.

• The acid value and water content were determined via a potentiometric (AOCS Cd 3d-63) (Metrohm, 876 Dosimat Plus) and coulometric (AOCS Ca 2e-84) (Mitsubishi Chemical Analytech, CA-31 moisture meter) titration, respectively. The density was measured at different temperatures, ranging from 20 to 70 °C, according to ASTM D7042 (Anton Paar, SVM3000). The coolant color was scored using the ASTM D1500 color scale (Haack Lange, Lico 690). The dynamic and kinematic viscosity were measured from 20 to 70 °C according to ASTM D445 (Anton Paar, SVM3000). The flash and pour point were determined according to ASTM D92 (Cleveland Open Cup) (NORMALAB, NCL 440) and ASTM D97 (ISL, CPP 5Gs), respectively. The coolant odor was assessed using EN 13725 (olfactometry). The total sulfur content in the coolant was quantified by ultraviolet fluorescence according to ASTM D5453.

• The specific heat capacity and thermal conductivity were determined according to ASTM E1269 (Mettler Toledo, DSC3) and ASTM D7896 (Flucon, LAMBDA), respectively. The temperature ranged from 20 to 70 °C.

• The electrical resistivity (20 Hz) and dielectric constant (relative permittivity, 100 kHz) were measured at 25 °C according to DIN 51111 (Flucon, Epsilon+). The dielectric breakdown was determined according to IEC 60156 (2.5 mm gap) using a 150 mL cell without a stirrer and with spherical electrodes (Megger, OTS100AF).

• The renewable carbon content and biodegradability were determined according to ASTM D6866 and OECD 301B, respectively.

2.1.2. Material Compatibility

Electrical wiring is typically insulated using PVC. Hence, a crucial first step in obtaining high-performance dielectric coolants is studying the PVC–liquid interactions. In that respect, a male USB A to male USB A 3.0 cable (RS PRO, 182-8844), in which the internal wires were enclosed in a blue PVC sheath, was selected. A single-component material compatibility test was performed.

The cable with an overall cable diameter of 5 mm was cut into pieces. Each piece should have a length of approximately 10 cm and a corresponding weight of approximately 4.0 g. Three cable pieces were immersed in 150 mL of L759 using a 200 mL closed glass vessel. The vessel was put into an oven at 80 °C for 14 days.

The PVC–L759 interactions were assessed by

• 1.any visual observations (e.g., changes in color);
• 2.quantifying the changes in cable dimensions, i.e., the length (ΔL %) and diameter (ΔD %); and
• 3.determining the acid value (AOCS Cd 3d-63), electrical resistivity (DIN 51111, 25 °C), and dielectric breakdown (IEC 60156) for used L759.

2.2. Functional Accelerated Aging

2.2.1. Thermal Accelerated Aging

2.2.1.1. Sample Preparation

500 mL closed glass bottles were used as recipients. Each recipient was filled with 450 mL of unused L759. Apart from the dielectric coolant, the following use case materials were added to the recipient to mimic actual liquid–material interactions:

• Heat hose segments

  2 pieces, approximately 1.7 g per piece (10 cm length, 2 cm width)

• O-ring seals

  2 pieces, approximately 1.5 g per piece (3.5 cm inner diameter, 0.35 cm thickness)

• Printed circuit boards

  5 pieces, approximately 2.3 g per piece (3.2 cm length, 2.2 cm width)

• Male USB A to male USB A 3.0 cable segments

  2 pieces, approximately 3.8 g per piece (10 cm length, 0.5 cm outer diameter)

• Ethernet cable segments

  2 pieces, approximately 4.9 g per piece (10 cm length, 0.6 cm outer diameter)

• High-voltage cable segments

  2 pieces, approximately 11.5 g per piece (10 cm length, 0.6 cm outer diameter)

• 3-pin connector power cable segments

  2 pieces, approximately 4.6 g per piece (15 cm length)

• Power supply cable segments

  2 pieces, approximately 11.2 g per piece (10 cm length, 0.9 cm outer diameter)

These materials were carefully selected based on (i) their functional relevance to the use case, (ii) the material type/chemistry, or (iii) preliminary single-component compatibility data retrieved from a systematic screening of more than 30 use case materials. As for the latter selection criterion, only materials significantly affecting the material (>10% dimensional changes) and/or liquid (e.g., significant color change) characteristics were chosen for this study. Additionally, a blank sample was prepared. This sample did not contain any materials and was used to study how the temperature directly impacts the coolant characteristics.

2.2.1.2. Aging Conditions

The aging temperature is critical, as a too low temperature significantly extends the test duration (cf. Arrhenius predictive model), whereas a too high temperature may introduce secondary effects that are not representative for the use case. However, no clear guidance regarding the choice of aging temperature exists. IEC 61125 has been reported, although this oxidation method is characterized by a relatively high temperature (120 °C) and does not take into account material–liquid interactions.

Given the temperature tolerance of both liquid and use case materials, the aging temperature was set at 100 °C. The prepared recipient (with materials) and blank (without materials) were put in an oven at 100 °C (T _AA ) for six months (AAT). Given a (standard) chemical reaction rate of 2 (Q ₁₀) and an average operating temperature of 50 °C (T _OPS ), the chosen aging conditions reflect an in-service operating time (IOT) of 16 years.

$$\mathrm{IOT}=\mathrm{AAT}\times {Q_{10}}^{(T_{AA}-T_{OPS})/10}$$

2.2.1.3. Coolant Characterization

The physical (density, kinematic/dynamic viscosity, and flash point), thermal (specific heat capacity and thermal conductivity), and electrical (electrical resistivity, dielectric constant, and dielectric breakdown) properties of both aged samples were determined according to the methods described in the Experimental Section, section .

2.2.2. Functional Testing

2.2.2.1. Setup

The thermo-electrical performance of aged L759 was evaluated using a single-board computer (Raspberry Pi 4 Computer, Model B, 2GB RAM). The experimental setup relies on heat transfer via natural convection and consists of a Raspberry Pi with visualization and load software (s-tui 1.0.2), a monitor screen, a keyboard, and a mouse. The Raspberry Pi (85.6 mm × 56.5 mm) was immersed in 2 L of dielectric coolant contained within a rectangular, closeable (but not hermetically sealed) glass container (23 cm L × 8 cm W × 22 cm H); see Figure .

2.

Static functional testing setup: Raspberry Pi 4 Computer (Model B, 2GB RAM) immersed in dielectric coolant L759.

2.2.2.2. Protocol

The Raspberry Pi 4 Computer was powered on and put in the rectangular glass container without (air cooling, benchmark 1) or with dielectric coolant (benchmark 2, used blank or test samples), see further. The Raspberry Pi and associated cable connections must be fully immersed; see Figure . Subsequently, the load program was opened via the Command Prompt window. This software can stress the central processing unit (CPU, 64-bit quad-core Cortex A72 processor) of the Raspberry Pi: initial (5%) versus full (100%) power load. It also provides a visualization of the CPU temperature (°C), frequency (Hz), and power load (%) as a function of time; see Figure .

3.

Load program: CPU temperature (blue), frequency (purple), and power load (green) as a function of time. Raspberry Pi 4 Computer running without (A) and with load (stressed) (B).

The functional testing included four experiments:

• Benchmark 1: air cooling

• Benchmark 2: immersion cooling using unused L759

• Blank: immersion cooling using an aged blank sample (i.e., aged without any use case materials) to isolate thermal effects

• Test: immersion cooling using an aged test sample to study combined thermal and material–liquid interaction effects

The protocol comprises two steps: Raspberry Pi conditioning (without load) for 15 min, followed by the actual testing (with load, 100% stress) for 180 min. The CPU temperature (°C), frequency (Hz), and power load (%) were monitored as a function of time and compared to benchmark data. The critical CPU temperature was set to 80 °C. Between tests, the Raspberry Pi was cleaned with unused L759. Prior to the next test, the performance of the cleaned Raspberry Pi was checked according to the “benchmark 2” protocol.

Following functional testing, a material compatibility review was performed. Specifically, the Raspberry Pi 4 Computer was examined for any material–liquid incompatibilities through visual assessment (compared with pictures before functional testing as reference). Particular attention must be paid to changes in material color, cable connectors, cable rigidity (stiffness), and the occurrence of corrosion.

2.3. Use Case

The use case concerns a precision cooled Ku:I2 V1 chassis as described in ref ; see Figure . The chassis is part of a 24U rack with a precision delivery manifold and liquid-to-air heat rejection.

4.

Cross-section of a precision cooled Ku:I2 chassis including a precision delivery manifold (1), server components (2), low-powered coolant pumps (3), a plate heat exchanger (4), and heat sink. Photograph courtesy of Iceotope Technologies Ltd. from ref . Copyright 2024.

The term “precision” refers to the immersion of specific electronic compounds rather than the entire hardware (cf. tank or “full” immersion). L759 is directly delivered to server hotspots (2) via an in-chassis manifold (1). Two micro pumps (3) create a (slow) recirculating coolant flow, enabling the heat to be transferred to a cooling distribution unit via a plate heat exchanger (4). Dry coolers reject the captured heat into the environment.

The Ku:I2 V1 chassis houses a 2.04 kW SR670 Lenovo server and contains 12.5 L of L759. A Raspberry Pi with artificial load software was implemented to stress the processors. Server and pump metrics were visualized using Grafana. Thermal and power data were collected from different integrated sensors over a period of 8 months. The physical, thermal, and electrical properties of L759 were also monitored. Each coolant sample was characterized as described in the Experimental Section, section .

3. Results and Discussion

3.1. Unused Dielectric Immersion Coolant

In 2023, Open Compute Project (OCP) IC published development guidelines for unused dielectric immersion coolants. The physical, thermal, and electrical properties of unused L759 were benchmarked against these recommendations, as shown in Table .

1. Full Property Analysis of Unused L759: Physical, Thermal, Electrical, Sustainability and Safety Aspects.

3.1.1. Physical Properties

Unused L759 has a kinematic viscosity of 7.3 mm²/s at 40 °C, supporting efficient heat transfer. The acid value and water content are 0.03 mg KOH/g and 160 ppm, respectively, both of which comply with the OCP IC development guidelines. Moreover, L759 is considered as fire safe, with a minimum flash point of 192 °C, exceeding the recommended threshold of 150 °C. The dielectric coolant has a pour point of −52 °C, pointing toward an excellent cold flow behavior. Additionally, the product does not contain any sulfur.

3.1.2. Heat Transfer Performance

In addition to density and viscosity, both the specific heat capacity and the thermal conductivity are crucial in assessing the thermal coolant performance. The density, dynamic viscosity, specific heat capacity, and thermal conductivity were measured as a function of temperature (20–70 °C); see Figure . Within this temperature range, the density and thermal conductivity are linearly dependent on temperature, whereas the dynamic viscosity and specific heat capacity exhibit a rather exponential behavior.

5.

Heat transfer properties of unused L759. (A) Dynamic viscosity (black ●), (B) specific heat capacity (green ▲), (C) density (blue ⧫), and (D) thermal conductivity (red ■) as a function of temperature, ranging from 20 to 70 °C.

Figures of merit (FOMs) were defined to represent the relationship among key thermal properties:

• Mouromtseff number (Mo)

• Intel FOM1: natural convection

• Intel FOM2: forced convection

• Intel FOM3: pressure drop

Mo is often used to assess the heat transfer rate of a coolant. Considering a coolant flow with a given velocity inside a fixed geometry at a given temperature, the dielectric coolant with the highest Mo has the highest heat transfer rate. This number is defined by the density (ρ, in kg/m³), the dynamic viscosity (μ, in mPa·s), the specific heat capacity (C _p , in kJ/(kg·K)), and the thermal conductivity (k, in W/(m·K)) of the dielectric coolant:

$$Mo=\frac{{\rho }^{\alpha }{k}^{\beta }{C_p}^{\gamma }}{{\mu }^{\delta }}$$

The exponents α, β, γ, and δ depend on the system’s heat transfer mode. Given the forced convection in the use case, these exponents are set at 0.80, 0.67, 0.33, and 0.47, respectively.

Mo was calculated across a temperature range of 20–70 °C; see Figure . A linear relationship was observed between Mo and temperature. Given the use case with an average operating temperature of 50 °C over 8 months, Mo is 36.2 for unused L759.

6.

Mouromtseff number of unused L759 in a forced convection regime at various temperatures (20–70 °C).

Apart from Mo, server OEM Intel has defined three FOMs inherent to 1P IC systems. The approach distinguishes between heat transfer modes: natural (FOM1) and forced (FOM2) convection.

FOM1 is derived from the Nusselt number and considers a natural convective heat sink with vertically oriented plate fins. The equation incorporates the thermal conductivity (k, in W/(m·K)), the thermal expansion coefficient (β, in 1/K), the specific heat capacity (C _p , in J/(kg·K)), the density (ρ, in kg/m³), and the dynamic viscosity (μ, in N·s/m²) of the dielectric coolant:

$$\mathrm{FOM}1=k{\left(\frac{\beta {\rho }^2{C}_p}{\mu k}\right)}^{0.2813}$$

As the thermal expansion coefficient of synthetic esters generally ranges between 0.0007 and 0.0008 (see ASTM D8240, Standard Specification for Less-Flammable Synthetic Ester Liquids Used in the Electrical Apparatus), FOM1 was calculated using a thermal expansion coefficient of 0.0008.

In contrast to FOM1, FOM2 relates to forced convection under laminar flow conditions. This equation includes the thermal conductivity (k, in W/(m·K)), the specific heat capacity (C _p , in J/(kg·K)), and the density (ρ, in kg/m³) of the dielectric coolant:

$$\mathrm{FOM}2=k{\left(\frac{\rho C_p}{k}\right)}^{0.3333}$$

Furthermore, in forced convection systems, pumps must overcome a certain pressure drop to maintain the coolant flow. This pressure drop directly correlates with the dynamic viscosity of the dielectric coolant (FOM3).

$$\mathrm{FOM}3=\mu$$

The Intel FOMs were evaluated at 50 °C for unused L759; see Table . Based on FOM1 (natural convection), L759 was categorized as a “Tier 2” dielectric coolant (FOM1 > 45). This indicates that L759 can cool both current and next-generation Intel processors. Also, unused L759 meets the performance criteria related to FOM2 (>19, forced convection) and FOM3 (≤0.015, pressure drop).

2. Figures of Merit Related to the Heat Transfer Performance of Unused L759.

3.1.3. Dielectric/Electrical Performance

To ensure electrical safety and mitigate the risk of short circuits, dielectric immersion coolants must exhibit both high electrical resistivity (≥2.0 GΩ·m) and high dielectric breakdown (≥35 kV). Unused L759 meets these electrical OCP IC development guidelines, with an electrical resistivity and a dielectric breakdown of 10 GΩ·m (at 25 °C, 20 Hz) and 77 kV, respectively. The dielectric constant is 3.5 at 25 °C (100 kHz).

3.1.4. Material Compatibility: Polyvinyl Chloride

PVC–L759 interactions were evaluated through a single-component material compatibility test conducted at 80 °C for 14 days; see Table .

3. Single-Component Material Compatibility Data: Changes in PVC Cable Dimensions (Length and Diameter) and L759 Properties (Acid Value, Electrical Resistivity, and Dielectric Breakdown).

The observed changes in cable dimensions (length and diameter) were considered as acceptable (<10%, compatible). The acid value, electrical resistivity, and dielectric breakdown changed upon testing; however, all were also within acceptable limits for used coolants.

3.2. Thermal Accelerated Aging

Both blank (L759 only) and test (L759, use case materials) samples were aged at 100 °C for six months. The effect of temperature on the coolant was studied via the blank, while the test sample accounted for the combined effect of the temperature and compatibility. The physical, thermal, and electrical properties of the aged samples were measured and compared to those of unused L759 (t = 0).

3.2.1. Physical Properties

The acid value, water content, and coolant color were evaluated after three and six months; see Figure . For both aged samples, the acid value and color increased over time, whereas the water content did not significantly change. The observed changes in acid value and color were more pronounced in the aged test sample than the blank, indicating a substantial impact of the use case materials. For the test sample, the acid value and color increased to 1.48 mg KOH/g and 3.2, respectively, after 6 months, while the increase was attenuated for the blank, reaching 0.82 mg KOH/g and 1.9 after 6 months. The flash point was 204 °C for the blank and 202 °C for the test sample after 6 months, indicating that aging did not compromise the coolant’s fire safety. For reference, unused L759 exhibited a flash point of 192 °C.

7.

Evolution of acid value (●), water content (▲), and ASTM D1500 color (■) in the blank (yellow) and test (blue) coolant samples over six months of aging. Lines are added to guide the eye.

3.2.2. Heat Transfer Performance

The density, dynamic viscosity, specific heat capacity, and thermal conductivity of the aged test sample were also measured over 20–70 °C and benchmarked against unused L759; see Figure .

8.

Heat transfer properties of unused (light) and aged (test sample, dark) L759: (A) dynamic viscosity (black ●), (B) specific heat capacity (green ▲), (C) density (blue ⧫), and (D) thermal conductivity (red ■) as a function of temperature, ranging from 20 to 70 °C.

The qualitative trends in thermo-physical properties remained consistent between the unused coolant and the aged test sample. The density and thermal conductivity exhibited a linear temperature dependence, whereas the dynamic viscosity and specific heat capacity demonstrated an exponential behavior. In terms of absolute values, changes in the density and thermal conductivity were negligible. In contrast, the dynamic viscosity showed a slight increase: 4.85 mm²/s (unused) versus 5.13 mm²/s (test) at 50 °C. The impact of aging on the specific heat capacity was more pronounced, particularly for the values in the low-temperature range: 1.95 kJ/(kg·K) (unused) versus 1.85 kJ/(kg·K) (test) at 20 °C and 1.99 kJ/(kg·K) (unused) versus 1.95 kJ/(kg·K) (test) at 50 °C.

The FOMs related to the heat transfer performance of unused versus aged (test) L759 are given in Table .

4. Figures of Merit Related to the Heat Transfer Performance of Unused versus Aged (Test) L759.

Given the minor changes in dynamic viscosity and specific heat capacity, the Mo (heat transfer rate) and FOM1 (natural convection) only slightly decreased upon aging: from 36.2 to 35.3 and from 56.2 to 55.5, respectively. FOM1 remained above the “Tier 2” threshold of 45, confirming that the aged test sample still qualifies as a “Tier 2” dielectric coolant. Interestingly, for IC systems operating under forced convection (FOM2), the impact of aging on the heat transfer performance of L759 was negligible. Additionally, the aged test sample met the pressure drop criterion (FOM3): 0.0051 N·s/m².

3.2.3. Dielectric/Electrical Performance

Both temperature and the presence of use case materials clearly affected the electrical resistivity of L759; see Figure . After six months of aging, the electrical resistivity decreased from 10 GΩ·m (t = 0) to 2.7 GΩ·m for the blank and further to 0.3 GΩ·m for the test sample. Remarkably, for the aged test sample, the electrical resistivity dropped to 0.4 GΩ·m after three months and remained nearly constant thereafter: it was 0.3 GΩ·m at six months, indicating an equilibrium was reached. In contrast, the dielectric constant of both aged samples did not significantly change upon aging: 3.5, 3.4, and 3.6 at 25 °C for the unused L759, blank, and test samples, respectively.

9.

Evolution of electrical resistivity (●, at 25 °C), dielectric constant (▲, at 25 °C) and dielectric breakdown (■) in the blank (yellow) versus test (blue) coolant samples over six months of aging. Lines are added to guide the eye.

Interestingly, the dielectric breakdown of the blank (76 kV) did not differ from that of unused L759 (77 kV), whereas a moderate decrease in dielectric breakdown was observed for the aged test sample (56 kV). This suggests that the decrease in dielectric breakdown is driven by the added use case materials rather than temperature alone.

3.2.4. Coolant Lifespan Estimation

OCP IC published guidelines outlining the specifications of used dielectric coolants in 1P IC applications. The document emphasizes important physical (flash point) and electrical (electrical resistivity and dielectric breakdown) threshold values.

In similar applications, often a specification is set for the acid value due to the risk of system malfunction associated with copper corrosion. ASTM D8240, for example, covers specifications inherent to less-flammable synthetic esters in an electrical power apparatus. For used liquids, a critical acid number was defined based on IEC 61125: ≤0.30 mg KOH/g. Exceeding this limit may accelerate degradation due to the autocatalytic nature of acid formation, thereby increasing the probability of system malfunction.

3.2.4.1. Physical Properties

OCP IC recommended that used coolants maintain a flash point of at least 150 °C. The aged test sample exhibited a flash point of 202 °C, indicating that L759 can guarantee long-term fire safety under actual operating conditions.

ASTM D8240 reported a critical acid number. For L759, the acid value exponentially increased over the time interval considered, as shown in Figure . Following the trendline for the acid value of the test sample (blue), a critical acid number of 0.30 mg KOH/g (y _test ) corresponds to an aging time (AAT, x) of 2.6 months. Applying the Arrhenius law, this translates to an estimated coolant lifespan of approximately seven years.

10.

Evolution of the acid value (●) of blank (yellow) and test (blue) coolant samples over six months of aging. The critical threshold value of 0.30 mg KOH/g (ASTM D8240) is indicated by a red line. Lines are added to guide the eye.

The acid value depends on both (combined) thermal and material effects and was therefore selected as a (physical) lifespan indicator.

3.2.4.2. Heat Transfer Performance

Based on the thermo-physical property analysis presented in section , the specific heat capacity was selected as a (thermal) lifespan indicator. However, to the best of our knowledge, thermal threshold values have not been defined to date. For this reason, functional test data are listed in the Results and Discussion, section .

3.2.4.3. Electrical Performance

Both electrical resistivity (0.3 GΩ·m) and dielectric breakdown (56 kV, 2.5 mm gap) of the aged test sample complied with the OCP recommendations for used liquids (≥0.2 GΩ·m, ≥6 kV/mm). Applying the Arrhenius law to these experimental data, the lifespan of L759 was set at 16 years.

Aiming at a high measurement accuracy, the electrical resistivity was selected over the dielectric breakdown as a­(n) (electrical) lifespan indicator.

3.3. Functional Testing

To date, thermal thresholds have not been established in the scientific literature. Also, for the data retrieved from aging experiments, it remains uncertain whether aged coolants can provide efficient cooling to electrical componentry, as the current OCP threshold values lack scientific validation for 1P IC applications. To address this gap, the thermo-electrical performance of the aged test sample was assessed through functional testing.

The aged test sample was used to cool a Raspberry Pi 4 Computer by immersion. Its performance was benchmarked against air cooling (benchmark 1) and immersion cooling using unused L759 (benchmark 2). The CPU temperature (°C), frequency (Hz), and power load (%) were monitored for 180 min; see Figure .

11.

CPU temperature as a function of the operating time when being cooled by air (red), by immersion in unused L759 (green), and by immersion in aged L759 (test sample, blue).

3.3.1. Thermo-Electrical Performance

For air cooling, the CPU reached its critical temperature (80 °C) after 3 min at full power load. The functional testing was ceased.

For immersion cooling, the CPU temperature was well-controlled (i.e., kept below 80 °C) at stress, ranging between 44 and 51 °C. This clearly demonstrated the benefits of immersion over air cooling. The steady rise in temperature is inherent to the static heat transfer regime in the glass container. Given a standard deviation of ±2 °C, the cooling performance of the unused L759 did not significantly differ from that of the aged one. Moreover, the CPU frequency (1800 Hz) did not change at full power load. No CPU malfunctioning was observed upon testing, indicating that the thermo-physical properties (density, dynamic viscosity, specific heat capacity, and thermal conductivity) of the aged test sample can account for efficient CPU cooling and that the critical acid number, as defined in ASTM D8240 (0.30 mg KOH/g), rather acts as a risk indicator for system failure than reflecting system failure per se. Based on these experimental performance data combined with the Arrhenius law, the coolant lifespan was estimated at 16 years.

3.3.2. Material Compatibility Review

The used Raspberry Pi 4 Computer was visually inspected and compared to its initial state; see Figure .

12.

Material compatibility review. Raspberry Pi 4 Computer before (pristine, A) and after (used, B) functional testing using an aged L759 (test sample).

No changes in material characteristics were observed with the naked eye. None of the PCB components became loose upon testing. No discoloration was observed. The cable connectors were easily unplugged after testing. Also, the immersed cables demonstrated low rigidity. No traces of corrosion were found. In addition, consecutive functional testing with unused L759 confirmed the good functioning of the Raspberry Pi 4 Computer.

3.4. Use Case

The key physical, thermal, and electrical properties of the aged test (100 °C, 6 months) versus use case (50 °C, 8 months) samples are shown in Table .

5. Key Physical (Acid Value), Thermal (Specific Heat Capacity), and Electrical (Resistivity and Dielectric Breakdown) Properties of L759: Aged Test (100 °C, 6 Months) versus Use Case (50 °C, 8 Months) Samples.

Interestingly, the acid value was significantly higher for the aged test than the use case sample: 1.48 and 0.08 mg KOH/g, respectively. In contrast, the specific heat capacity and electrical resistivity showed no significant difference between both samples. Both parameters reached steady-state values over time.

The dielectric breakdown did not significantly differ for the aged and use case samples, confirming that this parameter strongly depends on material–liquid interactions rather than on the aging temperature alone.

For these reasons, the acid value is identified as the sole primary lifespan indicator for L759. Also, the proposed aging methodology accurately reflects the actual material–liquid interactions in the use case.

4. Conclusions

This work reported on a comprehensive, system-oriented aging methodology to evaluate the long-term quality of dielectric immersion coolants and predict their lifespan in 1P IC applications.

The proposed approach integrated material compatibility, thermal stress, and functional performance factors. It offered a rigorous and reliable framework for evaluating the longevity of dielectric immersion coolants in 1P IC applications, thereby minimizing safety risks and the likelihood of system failures during expensive, long-term use case tests.

The method was validated using a biobased and nonhazardous synthetic ester coolant: L759 (Oleon NV). Use case data were derived from a precision immersion cooled Ku:I2 V1 chassis.

• Coolant Performance and Safety Metrics

  L759 demonstrated efficient heat transfer as reflected by its high Mo and thermal FOMs 1–3, all of which exceeded established threshold values. The coolant also met key environmental, electrical, and fire safety criteria. It exhibited an electrical resistivity of 10.0 GΩ·m, a dielectric breakdown of 77 kV, and a minimal flash point of 192 °C. Additionally, the coolant showed minimal interactions with PVC, a widely used cable insulation material. OCP property recommendations were met.

• Thermal Accelerated Aging

  Due to the careful selection of use case materials, the aging methodology (100 °C, six months) adequately simulated actual material–liquid interactions. Based on thermal aging and use case data, the acid value (physical; primary), specific heat capacity (thermal; secondary), and electrical resistivity (electrical; secondary) were identified as crucial lifespan indicators.

• Functional Testing

  Immersing a Raspberry Pi 4 Computer in aged L759 for 180 min demonstrated effective CPU cooling, as reflected in thermo-physical and dielectric/electrical property analyses. No material incompatibilities were observed.

Furthermore

• Ester-Based Dielectric Immersion Coolant

  L759 can comply with the lifespan of current (five years) and next-generation (seven, up to 10 years) servers. Hence, ester technology shows great potential in operating single-phase immersion cooling systems.

• Aging Methodology

  This work contributes to ongoing standardization efforts in 1P IC, promoting the definition of scientifically validated performance criteria. The principles about use case-driven material selection, thermal aging, coolant quality assessment (theoretical and experimental), and corresponding lifespan prediction are broadly applicable. The approach can be extrapolated not only to alternative coolant chemistries such as hydrocarbons but also to adjacent application domains where dielectric liquids are used, including batteries, electro-hydrodynamic pumps, and transformers.

Future research should prioritize the following areas:

• 1.Long­(er)-Term Performance Evaluation in Use Case

  To evaluate the performance of IC systems in HPC environments over five and up to 10 years. This will provide crucial insights into server reliability and the system’s total cost of ownership.

• 2.Advanced Coolant Formulations

  To investigate advanced coolant formulations aimed at, for example, further enhancing the performance on a system level.

• 3.Signal Integrity

  To investigate how IC impacts high-speed data transmission in electrical components such as interconnects and graphical processing units.

The authors declare no competing financial interest.