Applications, Design Methods, and Challenges for Additive Manufacturing of Thermal Solutions for Heterogeneous Integration of Electronics

Additive manufacturing (AM) enables new form factors and integration approaches for next generation electronics. An essential aspect for any electronics or photonics package is thermal management. Much research has been directed toward proper control of heat transfer within the package which is critical to assure device temperatures remain within allowable limits and to assure reliability of the package over its intended lifetime. Therefore, in conjunction with this effort, this article presents a recent perspective on applications, design optimization methods, and future challenges for AM of thermal solutions for heterogeneous integration of electronics.

A primary thermal application of AM of electronics (AME) is the rapidly expanding area of heat sink, cold plate, or heat exchanger fabrication. An extensive recent survey has been published examining common metal printing methods, postprocessing, heat transfer effects, and heat transfer augmentation approaches [1]. Select relevant methods are outlined in this article for metal AM with an eye toward electronics integration. AM of additional (i.e., polymer, ceramic, and composite) materials is further covered in support of greater package integration of comprehensive cooling solutions. Related to this material distinction, herein we may consider heat conducting elements as those components that benefit from high thermal conductivity versus fluid flow elements that do not necessarily require high thermal conductance. Beyond the surveyed range of additive methods, recent design and optimization approaches for heat sinks and cold plates (i.e., coolers) for electronics are also briefly examined. On this point, design of heat exchangers for AM is another adjacent, rapidly growing, and impactful research field [2], where three-dimensional (3D) printing unlocks new design freedom for thermal management of devices. Overall, since this article is focused on existing state-of-the-art fabrication methods and design techniques, the research and development of new material sets and processes is only briefly covered in the context of TIMs. Conceptual workflows for heterogeneous integration of electronics with AM thermal solutions are provided to stimulate future research in the field. The article concludes with a discussion of key opportunities and growth areas for thermal applications based on current fabrication and material limitations. Important heterogeneous integration packaging challenges for electronics are also identified as areas for greater work in the field over the next decade.

In this section, we provide a concise categorized review of metal, ceramic, and polymer AM processes aimed toward thermal management applications for electronics and copackaged optics. The role of each of these different material fabrication processes for heterogeneous integration of electronics in terms of thermal-fluid performance benefits is highlighted in Secs. 2.1–2.3, and key process characteristics are summarized in Table 1. Some specific discussion of heterogeneous integration is then provided.

Due to high thermal conductivity, k (W/m K), metals including aluminum (Al, $k∼130−190$ W/m K) and copper (Cu, $k∼350−400$ W/m K), plus associated alloys, are thermal conductors that are commonly used for fabrication of air-cooled heat sinks plus single and two-phase liquid-cooled cold plates for electronics. The following different metal AM processes enable the realization of various complex heat exchanger internal geometries with different materials at various structural feature length scales.

Selective laser melting (SLM) of metals is a process whereby powder metal that is distributed in a layer is selectively melted using a laser, and the part is built up layer-by-layer. As reviewed in [3], this process is now commonly used to fabricate heat sinks, and an example air-cooled heat sink for electronics is shown on the furthest left versus a range of conventionally machined heat sinks to the right in Fig. 1 [4]. The SLM process can achieve feature sizes as small as 200–400 μm with tolerances of the order of 50–250 μm [5]; see Table 1. Example materials for use in this additive process include tool steels, stainless steel (SS), AlSi10 Mg, AlSi12, Inconel, Ti64, tungsten, molybdenum, and Cu, to name a few. Key functional considerations for this well-established process include finished part material porosity and surface roughness which are important aspects in thermal-fluid applications since they relate to material thermo-physical properties, possible two-phase heat transfer (e.g., refer to Fig. 2), disruption of thermal boundary layers, and flow resistance within a heat sink or cold plate (i.e., heat transfer versus pressure drop tradeoffs) [1,4,6]. There are further processing considerations depending upon the material. For instance, typical laser power for 6061 Al is in the 50–100 W range with scan speeds of 100–400 mm/s; here, slower speeds plus greater power produce higher, $∼90%$⁠, relative density [7]. For AlSi12, near 90% relative density can be achieved at a scan speed of 200 mm/s using 100 W of laser power [7]. For Cu, 300–1000 W of laser power at scan speeds of the order of 300 mm/s may produce parts at > 95% relative density [8,9]. Pure Cu also has low laser power absorption, so alternative laser types or alloying additives may be needed [10]. Leftover powder removal can also be a postprocessing issue for complex internal channel geometries (e.g., see the multipass cooler for electronics in Fig. 3) and appropriate design considerations are required [11,12]. As for the final part, the most common approach to integrating a heat sink manufactured using SLM, or other energy beam methods like electron beam melting (EBM) or direct energy deposition (DED) [13], into an electronics package is to utilize a separate conventional TIM such as a grease or compliant gap pad layer between the heat sink and device or package.

The binder jet or Metal Jet [13–15] process for fabrication of metal heat sinks or cold plates is somewhat analogous to the SLM process above, where metal powder is bound by a binder. One of the key limitations of the SLM process is the metal weldability, and for binder jetting since no melting and solidification is involved, it opens up opportunities to some additional metals and alloys where sintering processes are effective. Here, metal powder is first prepared and loaded for printing. The powder is then spread into a layer and a binding agent is selectively jetted over the powder layer. This process is repeated layer-by-layer until a green, weakly bound, part is formed within the powder bed. The powder bed is then heated for the evaporation of liquid binder components and curing of the binder resulting in a high strength green part. Depowdering and cleaning then follow, and the green part is ready for debinding and sintering treatments. Sintering treatment causes the green part to reduce porosity and increase in density, up to > 95% density, depending upon the metallurgy and sintering process. The sintered part is typically ready for application; however, the part may be subjected to postprocessing treatments like hot isostatic pressing (HIP) to further densify the part close to its theoretical density as needed. Build speeds of 1990 cc/hr have been reported [16]. A TIM may be used for integration of a discrete Metal Jet heat sink, such as the topology optimized design from Ref. [17] shown in Fig. 4, or cold plate into an electronics package, Table 1.

An advantage of this process is that it may be adapted and extended to multimaterial builds for possible heterogeneous integration of thermal management components into an electronics package. A challenge with this approach lies in the postprocess sintering step, as temperature sensitive electronic active and/or passive devices may need to be integrated into the package. Thus, a stepwise manufacturing process may be needed where a portion of a multimaterial, i.e., heat sink plus substrate, package is prefabricated and sintered with active devices then mounted to the package postsintering. Other advantages include not needing printed supports (i.e., SLM may require support structures for part heat transfer features during manufacturing) and relatively fast print speeds.

The bound powder extrusion (BPE) process is a combination of metal injection molding (MIM) and fused deposition modeling (FDM), whereby metal powder is bound in a plastic matrix filament and then printed in a similar fashion as a pure polymer filament [18,19]. Thus, there is no loose powder or lasers used in the printing process. After printing, the metal part is (in some cases) washed to debind one binder-component and then heat treated to remove additional binder and sinter for achieving the final functional part. Challenges for this approach include achieving high final part density (more porosity initially occurs within bound powder extrusion parts given the extrusion process) and additional design rule requirements given the binder removal process. Build speeds of 16 cc/hr have been reported [19]. Representative washing and sintering steps are separately listed to each take a half of a day or more time for some processes [18]. Common materials and example postsinter feature sizes are reported in Table 1. Once more, a standard TIM may be used for integration of a discrete metal heat sink or cold plate into an electronics package.

The electro-chemical additive manufacturing (ECAM) process is a layer-by-layer additive approach based on the electrodeposition of metal. The typical electro-chemical additive manufacturing process is shown on the left in Fig. 5 [20]. Given a part geometry, a water-based feedstock flows across a printhead followed by micro-electrode printhead activation to form the print material on a vertically translating stage. The part is then printed upside down through precise pixel activation that drives a focused electric field to precisely deposit metal onto a base substrate. Part heights of the order of 10 mm at a build rate of 1–2 mm/hr (in production) are possible to print on a range of substrate materials including Cu, silicon (Si), alumina, and FR-4 (see Table 1) opening a range of potential electronics integration approaches for novel microscale thermal management structures [20,21]. This additive process can facilitate intricate geometries, Fig. 5 (right side image), and further exploits the complexity of topology optimized channels, fins, and/or porous structures that can benefit single and two-phase heat transfer at microscales for power dense applications.

Investment casting (IC) has been uniquely coupled with stereolithography (SLA)-assisted additive processes for fabrication of positive part resin patterns to produce eventual plaster casting negative molds then used to realize cast metal heat sink parts, as illustrated in Fig. 6. In Ref. [22], this workflow enabled the fabrication of complex heat sink designs for light emitting diodes (LEDs). The designs in [22] were determined using topology optimization for natural convection in air, where like Ref. [4], nonstandard heat sink fin geometries for enhanced thermal-fluid performance become available through a novel manufacturing process. As indicated in Table 1, a limitation of the SLA-assisted investment casting method might include minimum feature size [23], although tight fabrication tolerances using a broad array of metals is likely accessible. TIMs are once more commonly used for heat sink integration.

The use of polymers in AM of thermal management solutions for electronics is somewhat limited due to the low thermal conductivity (⁠$k∼0.2$ W/m K) of most base polymers. As explained in Sec. 2.2.1, FDM is an important approach to fabricate fluid flow directing manifold parts where high thermal conductivity is not critical; note that other AM processes have also been demonstrated [24,25]. Numerous investigations have also been performed on the AM of polymers with fillers for thermal conductivity enhancement in composite and heat sink parts, as briefly outlined in this section below, which represents a growth area. An excellent comprehensive review of AM in the context of polymers and polymer composites for heat exchangers is also provided by Deisenroth et al. [26], where additional fabrication methods are outlined, and a broad range of representative print speeds (depending on printing apparatus and material) is provided in Ref. [27].

Fused deposition modeling (FDM) is a logical choice for the AM of polymer manifolds that can provide an engineered flow path for liquid or air delivery to an electronics cold plate or active/passive devices. As explained in different studies [28,29], this design approach may provide an integration opportunity for cooling solution modularity, lightweighting, and multimaterial assembly that enables the retention of high thermal conductivity parts in locations of greatest thermal importance (e.g., closest to the electronics package); see Fig. 7. In terms of electronics package integration, polymers also allow for the design of unique (e.g., snap fit) connections that further eliminate weight associated with mechanical fasteners. Table 1 provides some representative FDM printing characteristics from the literature [30]. Recently, FDM has been used for additive fabrication of polymer heat sinks (with print speeds of 25 mm/s) in Ref. [31] using Ice9 Rigid Nylon from TCpoly [32]. Compared with a conventional Al part, reference straight fin polymer heat sinks are reported to have a ∼2.5× higher total thermal resistance, $Rth_tot$⁠. Additionally, the material thermal conductivity is reported in Ref. [31] to be anisotropic with different printed in-layer, $k∼3.23$ W/m K, and cross-layer, $k∼0.88$ W/m K, values. These anisotropic material properties are employed in design optimization to realize a 10–20% enhancement in performance relative to the straight fin polymer heat sink reference design. Further advancements in the thermal conductivity of the base polymer material itself may offer future advantages for high performance and lightweight heat sinks for electronics. Moreover, unique polymer-to-metal direct attachment capabilities, as described in Ref. [33], suggest interesting heterogeneous integration directions for future electronics once a polymer part has been printed.

Filaments composed of various polymers such as PLA (polylactic acid), ABS (acrylonitrile butadiene styrene), ASA (acrylonitrile styrene acrylate), PC (polycarbonate), PPS (polyphenylene sulfide), PEEK (polyetheretherketone), and others, typically exhibit excellent thermal insulating properties and negligible heat conduction. Despite their advantageous traits like reduced weight and simplified processing compared to metal counterparts, there has been a concerted effort to enhance the thermal conductivity of polymers for 3D printing by integrating thermally conductive fillers [26]. These fillers may include metallic fillers (e.g., Al, Cu), carbon-based fillers (e.g., graphene, carbon nanotubes (CNTs)), or ceramic fillers (e.g., boron nitride (BN), aluminum oxide (Al₂O₃ or alumina), aluminum nitride (AlN), and silicon carbide (SiC)), among others [34–39]. Interestingly, heterogeneous thermal composites with inclusions aligned in the direction of the polymer melt flow (through the additive nozzle) during printing [40] are conceivable. Also, external physical fields [41] (e.g., a magnetic field) may be used with functionalized fillers to create a custom anisotropic thermal composite. While metallic and carbon-based fillers exhibit high thermal conductivity, their electrical conductivity poses limitations in various insulation applications. Conversely, ceramic particles offer excellent electrical insulation properties alongside high thermal conductivity, rendering them suitable filler materials for numerous electronic TIM applications. FDM [42], as summarized in the prior Sec. 2.2.1 and Table 1, and similar direct ink writing (DIW) [43] methods stand out as two of the most widely employed material extrusion/deposition AM techniques for fabricating polymer composite-based thermal management components, owing to their capability to process a broad spectrum of commercial thermoset matrices and nanofillers. Reports have indicated significant improvements in the thermal performance of certain thermal management components, such as heat sinks [44,45], made of anisotropic polymer composites.

Like polymers, ceramics can be employed for nonheat-conducting fluid delivery structures in electronics cooling solutions. Advantages of ceramics relate to weight reduction, corrosion resistance, and favorable thermal-mechanical properties such as a coefficient of thermal expansion (CTE) that is more closely matched to semiconductor device materials such as Si or SiC.

Somewhat analogous to metal AM processes that use MIM powder in a slurry with an optical printer (e.g., see Ref. [46]), lithography-based ceramic manufacturing (LCM) is a fabrication method that involves a ceramic-loaded liquid slurry into which a movable build platform is dipped from above [47,48]. This platform is selectively exposed from below to a layered image in visible blue light via a projection system. The process is repeated layer by layer to create a 3D green ceramic part that is thermally postprocessed and sintered resulting in a fully dense part. This procedure allows macroscale porous lattice structures to be fabricated out of materials such as Al₂O₃ or AlN [49] with micron-scale struts and features, Table 1. These lattices can have complex designs in 3D similar to the Cu-based manifolds proposed by Palko et al. [50,51], where the lattice serves as a wick-like structure, Fig. 8 right side image, that may be integrated into electronics cooling solutions for possible capillary fluid delivery to a cold plate. The part shown in Fig. 8 is fabricated using LithaLox 360, a 99.8% alumina ceramic material (related to that in Ref. [52]), on a CeraFab S65 printer that has reported build speeds up to 150 layers per hours at 10–200 μm layer thickness [48].

As described in the literature and online [53,54], selective laser reactive sintering (SLRS) builds from selective laser sintering technology, where a reactive gas is introduced into the additive process to realize nonoxide ceramics. This relatively new technology is of interest for electronics package-integrated cooling [54] that may allow for elimination of a TIM between the package and cooling solution. Some possible material sets for AM are listed in Table 1, for reference. Reported laser power for selective laser reactive sintering is in the 4–5.25 W range with scan speeds of 100 mm/min [53].

In Eq. (1), the conductive thermal resistances from the device up through the heat sink are summed across all, i = 1 to j, layers and $Rth_cnv$ is the convective thermal resistance from the heat sink to the air. The summation of the conductive thermal resistance terms is then expanded further in Eq. (2) to include the conductive thermal resistance of the device, $Rth_d$⁠, TIM1 layer, $Rth_TIM1$⁠, metal lid, $Rth_lid$⁠, TIM2 layer, $Rth_TIM2$⁠, and heat sink, $Rth_hs$⁠. The majority of the efforts related to the AM processes outlined in Sec. 2 revolve around 3D printing discrete thermal management components, such as heat sinks or cold plates, which are subsequently mounted on to electronic devices using prefabricated TIMs. These efforts are associated with the last two heat sink related terms (⁠$Rth_hs$ and $Rth_cnv$⁠) in Eq. (2). Specifically, new heat sink or cold plate geometries enabled by AM allow for modification of the fin efficiency, η, convective heat transfer coefficient, h, or surface area, A_s, of the structure. On the other hand, TIMs are designed to fill microscopic gaps or roughness imperfections in the mating surfaces where electronic devices meet heat dissipation components. While the purpose of a TIM is to minimize the conductive thermal resistances (e.g., $Rth_TIM1$ and $Rth_TIM2$⁠) in a package, these resistances often remain as heat transfer bottlenecks. The thermal resistance of a TIM is proportional to its thickness, t, and inversely proportional to the layer thermal conductivity, k, and cross section area, A_xs; see Eq. (1). Electronics constraints, such as electrical isolation, may limit the thermal conductivity and minimum thickness of appropriate materials for a TIM.

Given the above understanding, in addition to increasing cooling performance of AM heat sinks and cold plates through new AM capabilities and design-for-AM strategies, emerging AM research is directed toward bridging the interface gap between the heat source or component to be thermally managed (i.e., electronics and potentially copackaged optics) and the heat sink (i.e, cooling components). This is particularly crucial for achieving greater miniaturization and higher integration levels for electronic systems, and examples including an overview of select case studies are provided in this section.

Owing to their exceptional electrical insulation and dielectric properties, printed TIM materials hold promise for future applications in electronics integration. Ultimately, a partial or even complete AM workflow for heterogeneous integration of electronics is likely to consist of composite TIMs and/or thermal solutions, as illustrated conceptually in the upper workflow image of Fig. 10. This type of process workflow may require postprocessing (e.g., thermal treatment, ultraviolet (UV) curing, application of pressure, application of external physical fields, etc.) of the polymer composite material and/or assembly to achieve the final desired material state or functional interfaces. Here, consideration of proper timing for package integration of active and passive devices in the process workflow is critical for device survival.

As a first case study, a 3D printed TIM made of a reduced graphene oxide/CNT pillar array infiltrated with polydimethylsiloxane (PDMS) matrix was recently developed, as reported in Ref. [55]. The DIW pillar array by itself was shown to have a thermal conductivity of 38.9 W/m-K. The full TIM was demonstrated to have a thermal conductivity of 6.04 W/m-K when the pillar array was embedded in the PDMS matrix. In Ref. [55], the TIM was printed and postprocessed (including freeze drying of the pillar array prior to applying PDMS) separately then applied to a heat source after fabrication. This type of printed TIM might easily be combined with discrete AM heat sinks, such as the air-cooled 3D printed AlSi12 design in Ref. [4], to realize a form of the upper workflow shown in Fig. 10 where both the TIM and heat sink are applied to an active device/package after postprocessing. The combined approach seeks to address the conductive thermal resistance terms, $Rth_TIM1$ and $Rth_TIM2$⁠, along with the heat sink conductive and convective thermal resistance terms, $Rth_hs$ and $Rth_cnv$⁠, respectively, found in Eq. (2) using the AM workflow.

Due to the significant thermal resistance posed by TIMs, recent reports have explored TIM-less cooling solutions facilitated by AM, aiming to integrate a fluid flow manifold directly onto electronic devices, thereby drastically reducing total thermal resistance. The reader is referred to the lower workflow image in Fig. 10. Among various AM techniques, SLA with polymers can be employed for fabricating complex, high resolution hollow structures and microfluidic systems [56].

Specifically, in a second case study [57], a fluid manifold was 3D printed using the SLA technique to provide jet impingement for direct fluid cooling of the backside of high-performance chips. The printed manifold is affixed to the electronics package substrate with O-rings for sealing. Following the lower leftmost workflow image in Fig. 10, the use of the 3D printed manifold for direct fluid jet impingement of the device in Ref. [57] enables the removal of the package lid and both TIMs (i.e., eliminating $Rth_TIM1, Rth_lid$⁠, and $Rth_TIM2$ in Eq. (2)), refer to Fig. 9, leading to a ∼50% reduction in the total unit (area-averaged) thermal resistance. In addition to this case study, a more aggressive one-step or lab-on-printed circuit board (PCB) approach was described in Ref. [58], involving the direct AM of a microfluidic manifold on to a PCB using SLA. Beyond, SLA and FDM, other AM processes can be used for fluid manifolds aimed toward direct chip cooling for package integration [24]. These approaches also affect the convective thermal resistance term, $Rth_cnv$⁠, in Eq. (2).

Regardless of technique, a significant challenge lies in matching the thermal-mechanical properties of a manifold and the electronics substrate on to which it may be printed, particularly the CTE. Developing photocurable composite formulations with a low CTE through resin doping with low-CTE solid fillers is one practical approach to address the issue of dissimilar material bonding [58]. Alternatively, a manifold (e.g., see Palko et al. [50]) might be “floated” over a cooler surface like that in Ref. [59] for a two-phase cooling configuration, allowing for fluid leakage and capture using other mechanisms, thereby completely avoiding CTE mismatch issues.

In addition to AM fluidic manifolds for direct liquid cooling of electronics, another approach involves directly 3D printing heat sinks or surface area enhancements onto electronics to circumvent the thermal resistance introduced by TIMs. Again, the reader is referred to the lower image in Fig. 10, where different workflow scenarios are outlined.

In our third case study [43], researchers demonstrated DIW to fabricate 14 wt% CNT/silicone heat sinks directly onto a chip attached to a circuit board, exactly following the lower middle workflow in Fig. 10. Radio frequency-assisted curing of the printed material outside of an oven was uniquely utilized. The heat sink was then shown to work in effectively lowering the device temperature by 1–3 °C when operating at 5% and 97% usage. One AM heat sink design is further shown to achieve the same steady-state device temperature performance when compared with an Al heat sink. Here, CNTs serve as conductive nanofillers and rheological modifiers, enhancing both thermal conductivities and the printability of the silicone inks. This innovative printing technique has the potential to eliminate the need for TIMs (again possibly eliminating one or more of the $Rth_TIM1$⁠, $Rth_lid$ and $Rth_TIM2$ terms in Eq. (2)) and the step of heat sink mounting, thereby offering an efficient means to manufacture high-performance polymer-based heat sinks with reduced production time and energy requirements.

Numerous other approaches for package integrated cooling via AM are being currently explored. Some federally-funded examples include projects related to topology optimized AM wicks for direct two-phase cooling and AM evaporators with” coral-shaped” heat sinks for immersion cooling of processors in data centers [60]. These AM strategies might further aggressively reduce the heat sink or cold plate convective thermal resistance (i.e., the $Rth_cnv$ term in Eq. (2)) for the final integrated electronics package.

Numerous advances in understanding of design-for-AM of heat exchangers is covered in a range of recent reviews [1–3,61,62]. Parametric geometry optimization is an effective tool to design structures which may be manufactured using subtractive manufacturing [63] or require the use of AM [64]. These methods are limited in design space coverage involving only geometric shape, due to explicit parameterization; however, they are typically compatible with many simulation tools since a body-fitted mesh can defined when using explicit geometry. AM constraints would need to be explicitly defined in the geometry parameterization and both gradient-based and gradient-free methods are applicable.

As an alternative to explicit parametric geometry, both discrete [65] and continuous [66] rule-based methods can be used to generate parametrically defined geometry. These generative parameterizations may expand the design domain to tune both structural shape and topology; however, structural variety is limited by the formulation of the generative model. Enforcing AM constraints using these methods may be challenging depending on the construction of the generative model. The generative parameterization may also require the use of gradient-free optimization algorithms.

Inverse design (e.g., topology optimization) techniques for the optimization of heat sink fin and cold plate channel geometry is being widely studied [2,61]. As shown in these reviews, the field of design optimization for AM of heat sinks and cold plates has advanced rapidly over the last ∼15 years with many numerical investigations and some experimental examples involving AM available in the literature; in particular, the reader is referred to Table 5 in Ref. [2] for a comprehensive summary related to topology optimization. A topology optimized design of a manifold microchannel (MMC) heat sink from Ref. [17], corresponding to the representative prototype part from Fig. 4, was designed for a power electronics module cold plate and is shown in Fig. 11, for reference. Here, we see in the zoomed view that topology optimization enables unique fin designs that may perform better than traditional structures. Figure 12 shows another topology optimized and additively manufactured MMC heat sink structure with a further nonintuitive fin design [67]. Beyond the range of examples found in the literature, some commercial software companies have also been formed recently [68–70] looking to capitalize on the trend in this field and bring the technology to bear in the market. These tools leverage ideas around generative design, start to separately explore the use of artificial intelligence algorithms (a field that is rapidly growing), and complement other existing commercial software resources that are well established for design optimization [71,72]. Ultimately, in the future, it is expected that this range of tools will be integrated into digital twin frameworks for complete thermal-mechanical package-to-system level co-optimization.

Based on the above review, growth areas and challenges for integration of additively manufactured thermal management solutions for electronics can be separated into three categories including the manufacturing itself, design-for-AM, and integration. The complexities in analyzing the cost of AM and process scalability are also discussed.

Properties of additively manufactured parts play an important role in several physics domains relevant for heterogeneous integration. The relevant physics, key physical parameters, and print characteristics have been summarized in Table 2. For heat conduction, the material porosity/relative density, layer thickness, infill or hatch pattern, anisotropy, composite filler orientation, and surface roughness play an important role in the thermal conductivity, k, specific heat capacity, C_p, the effective cross-sectional heat transfer area, A_xs, and thickness, t, of the part. Extending to thermal-fluid performance, part tolerances and feature sizes in addition will impact the permeability, K_e, cooled surface area, A_s, fin efficiency, η, and a variety of other factors captured in the convective heat transfer coefficient, h. The thermal-mechanical performance of a part is influenced by several print characteristics relating to the CTE, α, as the key parameter; see Table 2. Mechanics are also impacted by the full range of print characteristics shown in the table through changes to the Young’s modulus, E, density, ρ, and Poisson’s ratio, ν. While all of the print characteristics may influence the key physical parameters, those noted with a check mark in Table 2 are considered to play a more relevant role.

Thus, there are several manufacturing challenges and opportunities relating to thermal performance as summarized in the following list:

• Material porosity as it relates to thermal-mechanical (strength, surface roughness) properties of the final part,

• Realization of reliable interfaces for dissimilar materials; e.g., thermal-mechanical (thermal stress, thermal interface resistance, CTE) considerations for dissimilar material interfaces,

• Development of thermally conductive (low thermal resistance, high temperature capable) polymers and composites,

• Consideration of materials compatible with current and future electronics coolants,

• Development of fast metal AM processes that can achieve feature resolution in the range of 1–5 μm (e.g., for two-phase cooling solutions).

In many cases, 3D printed parts tend to be porous. Porous materials typically have lower thermal conductivity than fully solid materials, this results a lesser performance in heat conduction applications. Furthermore, such parts also tend to have surface roughness. When used in convective cooling applications, high surface roughness increases heat transfer performance, particularly for two-phase systems [73]. This comes at a cost in terms of reliability, where fluid may flow with higher resistance and degrade the part when passing over such textured surfaces. Therefore, controlling material porosity in the AM process to alleviate such concerns is a challenge and depending on the application (e.g., cold plate versus fluid/air flow manifold), the requirement for a water-tight design is different. Nonetheless, many porous material heat transfer applications exist, and porous materials may be well-used in air, single-phase (liquid), and two-phase (liquid–vapor) cooling solutions. Thus, being able to specify and spatially control porosity during fabrication is a major manufacturing opportunity which may be realized in the near term (1–3 years).

The properties of parts fabricated by AM also play a critical role in thermo-mechanical performance. Part porosity manifests in nonuniform feature sizes throughout the part, and surface-roughness facilitates the initiation of crack propagation [74]. The printing of dissimilar materials also affects thermo-mechanical performance due to CTE mismatch. Hence, multimaterial printing may further evolve in the midterm (3–6 years) to realize low conductive thermal resistance interfaces and/or composite TIMs. Here, optimal strategies are needed to mate materials (e.g., Cu to FR-4, ceramics, etc.) to realize continuous (nonporous) and high conductance thermal interfaces that are reliable. Thermal stress from the manufacturing process itself, in addition to actual use, should be considered which may require engineering of a thermal composite material CTE or interfacial properties. The implications of thermo-mechanical behaviors have been widely investigated for power electronics applications, and the reader is referred to Ref. [75] for more details. The end goal is to ensure a mechanically robust and reliable interface between an electronics platform and heat sink or cold plate.

The prior manufacturing challenges relevant to AM of thermal solutions for electronics may potentially be addressed as materials and processes evolve, and there has been great advancement in materials such as nanocomposites [76] and high entropy alloys [77]. Additionally, in the mid-to-long term (6–9 years) further research into reducing the minimum feature size below 10 μm for high-throughput printed metal parts may be achievable in order to fully connect to the requirements for advanced wick structures [78] found in high performance two-phase (e.g., boiling, vapor chamber, or capillary cooling) heat transfer applications for electronics. The use of such wick structures can connect well with package integrated fluid delivery and cooling schemes, Secs. 3.2 and 3.3, for compact, high performance near-junction thermal management solutions.

Major design method challenges for AM of heat sinks and cold plates include how to account for the following items in the design technique itself:

• Build orientation effects [1,31,79],

• Material porosity variations [1,5,51],

• Finished part surface roughness effects [1,12],

• Removal of powder from the final finished part (depending on print method) [12]

• Constraints on overhanging features [80],

• Composite material representations.

Additionally, from a computational perspective, major challenges also exist in relation to the design optimization for multiphysics thermal-fluid systems in 3D at the global heat exchanger scale. As described in Ref. [17], a periodic local unit cell approach may be employed to support a design problem that is computationally tractable. Alternatively, a homogenization approach may be exploited to optimize fluid flow or heat sink channel structures at the global scale using effective porous media material properties on a relatively coarse mesh with eventual dehomogenization of an explicit microstructure based on a range of numerical techniques [81,82]. The inverse design strategy using the homogenization method has the added built-in benefit of being able to effectively utilize AM material porosity effects for the design of new heat exchanger topologies [82], and utilization of the approach is expected to grow as an opportunity in the near term (1–3 years). Topology optimization for a myriad of flow types, as comprehensively reviewed in Ref. [83], including turbulent flow [84,85], and considering heat transfer [86], then adds significant additional complexity to the above discussion given the numerous different flow models available in the literature each with their own merit for various electronics cooling applications. As an example, the reader is referred to the range of computational fluid dynamics (CFD) approaches available to model fluid jet impingement [87], which is effective in electronics cooling [88,89]. Here, in the midterm (3–6 years), there may be opportunity to leverage physics-based machine learning methods (in the forward problem solution) to probe the intersection of heat exchanger AM and design for large-scale conjugate heat transfer under turbulent flow. Finally, design optimization for heterogeneous integration of embedded thermal solutions into power conversion, copackaged optics, and compute electronics packages considering electrical, optical, and thermal functions simultaneously is an enormous area of future research opportunity. Embedded structures related to near-junction cooling [90,91] and thermal metamaterials for heat flow control within the package [92–94] for ultracompact heterogeneous integration might be realized in the mid-to-long term (6–9 years). Here, functionally-graded thermal design using AM methods for local device hotspot mitigation becomes a possibility. Design of an integrated electronics cooling solution for conjugate heat transfer considering phase change of the working fluid in an optimized porous structure or wick [6,89] is another ambitious area for exploration that would compliment any advances in AM for precision micron-scale porosity, as suggested above. However, these various ideas require further development and maturity of multimaterial/composite (i.e., metal plus polymer plus ceramic) AM processes plus associated electronic design automation (EDA), co-optimization, and digital twin tools, where high performance cooling, continuous thermal interfaces and TIMs, routing for heat flow control, and fluidic design aspects are more holistically addressed.

Integrating AM thermal solutions with electronic systems involves several challenges. The first challenge is to ensure the durability and reliability of 3D printed components since they must be resilient under prolonged thermal stress, humidity, and potentially corrosive environments. To maintain cooling performance and structural integrity, the aging and degradation of materials, especially for polymer-based 3D prints, must be addressed to avoid compromising the cooling efficiency or structural stability over time. Second, enhancing the bonding strength at the interface between electronic components and directly integrated printed parts is critical to ensure a stable and reliable connection that can endure repeated thermal cycling without losing adhesion or showing signs of delamination. This issue is critical at polymer-to-metal (e.g., TIM-to-heat sink or PCB-to-cold plate) interfaces, and an example of this interface challenge is described in Ref. [95] in relation to the AM of a power PCB directly on a metal substrate (as a cold plate surrogate). Third, countermeasures must be implemented to prevent electrical short circuits when directly integrating 3D printed parts. Technical solutions may involve applying insulating layers or films to an electronic surface before 3D printing material on the surface. The design or positioning of the components themselves may be further adjusted (e.g., through embedding [96]) to ensure that different voltage potential regions are not exposed on the same surface and that these regions are indeed protected from unintended contact with AM metal parts. These electrothermal considerations are essential for the effective integration of AM-fabricated thermal solutions within electronics packages and require future research.

As described in a comprehensive and well-cited NIST report [97], the cost and scalability of AM are influenced by multiple factors that include raw material costs, printing machine/apparatus costs, production speed, labor, facility energy consumption, and part postprocessing requirements. These complexities necessitate a detailed, case-by-case cost estimation approach rather than a single formula. In many cases, the costs of AM exceed those of traditional manufacturing methods since AM does not benefit from economies of scale due to slow deposition rates and build capacity that may be limited [97,98]. Technologies like powder bed fusion (e.g., SLM) and DED are generally expensive but yield high-quality metal parts, making them suitable for complex geometry, low-volume part production. In contrast, material extrusion (e.g., FDM) and binder jetting are more cost-effective, especially for plastics, which supports their use in prototyping and larger batch production. While distributed manufacturing may reduce inventory and transportation costs with improvements in automation, high material costs currently limit widespread adoption. Scalability across AM technologies also varies significantly. SLM and DED are less scalable for mass production due to high costs and slower build rates, while material extrusion and binder jetting offer better scalability, particularly for polymer-based products. Generally, AM currently excels in small-batch or on-demand production. However, future advancements in automation and reductions in machine and material costs may enhance scalability for broader commercial electronics manufacturing applications.

In this article, we outlined the importance of AM in support of heterogeneous integration of thermal solutions for electronics. Fabrication of cooling solutions including cold plates, heat sinks, fluid flow manifolds, TIMs plus thermal composites was reviewed. Conceptual workflows for heterogeneous integration of such components and materials were proposed, and yet other approaches should still be conceivable. Trends in design-for-AM of electronics cooling solutions were additionally covered. Future directions were suggested for research and development of AM for electronics cooling with an aim toward heterogeneous integration. Finally, significant challenges in manufacturing, design, integration, cost, and scalability for AM were discussed, which represent opportunities for further research in this important and fast moving field.

This article is an expanded version of a section contained within the Institute of Electrical and Electronics Engineers (IEEE), Electronics Packaging Society (EPS), Heterogeneous Integration Roadmap (HIR) [99] chapter on Additive Manufacturing & Additive Electronics for Heterogeneous Integration. The HIR activities are sponsored by IEEE in coordination with the Electronics and Photonics Packaging Division (EPPD) of ASME. The authors would like to thank Dr. James Palko for insightful discussions regarding this work.

The authors attest that all data for this study are included in the paper.

Abbreviations

AM =

additive manufacturing

AME =

additive manufacturing of electronics

BPE =

bound powder extrusion

CFD =

computational fluid dynamics

CAM =

electro?chemical additive manufacturing

CTE =

coefficient of thermal expansion

DED =

direct energy deposition

DIW =

direct ink writing

EBM =

electron beam melting

ECAM =

electro-chemical additive manufacturing

EDA =

electronic design automation

EPPD =

electronics and photonics packaging division

EPS =

electronics packaging society

HIR =

heterogeneous integration roadmap

HIP =

hot isostatic pressing

IC =

investment casting

IEEE =

Institute of Electrical and Electronics Engineers

FDM =

fused deposition modeling

LCM =

lithography-based ceramic manufacturing

LED =

light emitting diode

MIM =

metal injection molding

MMC =

manifold microchannel

PCB =

printed circuit board

SLRS =

selective laser reactive sintering

SLA =

stereolithography

SLM =

selective laser melting

TIM =

thermal interface material

UV =

ultraviolet

Material Abbreviations

ABS =

acrylonitrile butadiene styrene

ASA =

acrylonitrile styrene acrylate

BN =

boron nitride

CNT =

carbon nanotube

FSLA =

free sintering low alloy

PEEK =

polyetheretherketone

PEKK =

polyetherketoneketone

PC =

polycarbonate

PLA =

polylactic acid

PPS =

polyphenylene sulfide

SS =

stainless steel

Symbols

α =

coefficient of thermal expansion, 1/K

ρ =

density, kg/m³

η =

Poisson’s ratio

ν =

fin efficiency

A_s =

surface area, m²

A_xs =

cross-sectional area, m²

E =

Young’s modulus, GPa

h =

convective heat transfer coefficient, W/m² K

k =

thermal conductivity, W/m-K

K_e =

permeability, m²

$Rth_cnd$ =

conductive thermal resistance, K/W

$Rth_cnv$ =

convective thermal resistance, K/W

$Rth_tot$ =

total thermal resistance, K/W

t =

thickness, m