Mass transfer techniques for large-scale and high-density microLED arrays

Inorganic-based micro light-emitting diodes (microLEDs) offer more fascinating properties and unique demands in next-generation displays. However, the small size of the microLED chip (1–100 µm) makes it extremely challenging for high efficiency and low cost to accurately, selectively, integrate millions of microLED chips. Recent impressive technological advances have overcome the drawbacks of traditional pick-and-place techniques when they were utilized in the assembly of microLED display, including the most broadly recognized laser lift-off technique, contact micro-transfer printing (µTP) technique, laser non-contact µTP technique, and self-assembly technique. Herein, we firstly review the key developments in mass transfer technique and highlight their potential value, covering both the state-of-the-art devices and requirements for mass transfer in the assembly of the ultra-large-area display and virtual reality glasses. We begin with the significant challenges and the brief history of mass transfer technique, and expand that mass transfer technique is composed of two major techniques, namely, the epitaxial Lift-off technique and the pick-and-place technique. The basic concept and transfer effects for each representative epitaxial Lift-off and pick-and-place technique in mass transfer are then overviewed separately. Finally, the potential challenges and future research directions of mass transfer are discussed.

These authors contributed equally to this paper. * Author to whom any correspondence should be addressed.
Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
High-resolution microLED displays rely on millions of polychromatic self-emissive elements that consist of red, green and blue (RGB) microLEDs [20]. Figure 1 shows the simplified fabrication process of microLED displays (center part in figure 1) and representative applications that have been described above. Generally, limited by the growth techniques, it is difficult to simultaneously grow RGB microLEDs on a epitaxial wafer (only ∼8 inches [21,22]). Thus, deterministic assembly of microLED chips from different growth/donor substrates is required. However, since the feature size of microLED chips are <100 µm, there is a trade-off between tiny feature size and fabrication feasibility. There are three enormous challenges. (a) The extreme transfer efficiency (∼tens of millions h −1 ) is vital because of their vast numbers. For an 8 K displays, more than ∼100 million chips need to be transfered onto the receiver substrate, which will take several weeks for traditional pick-and-place techniques (e.g. assembly throughput ∼8000 chips per hour [23][24][25] for current flipchip bonding equipments [23,26]). (b) The extreme placement accuracy is critical for microscale chips (∼5% of the microLED chip size [20,27]). For example, the transfer error of a 10 µm-sized microLED chip should be smaller than 0.5 µm, which is far beyond the accuracy of traditional transfer techniques (e.g. only ∼30 µm [28]). (c) The extreme reliability (∼99.9999% [21]) is also essential. Otherwise, a 0.01% failure rate could result in thousands of dead pixels [29]. In short, traditional assembly methods are not practical for MicroLEDs.
As a revolutionary technique specified by industry and mass transfer, which can release massive microscale chips (e.g. ∼millions h −1 ) from the donor/growth substrate an efficient rate and move them to the backplane/receiving substrate with high precision and reliability, it has proven to be a promising solution. These techniques typically introduce physical/chemical interactions (e.g. Van der Waals (VDW) force [33,34], fluid tension [35], electrostatic force [36,37], laser ablation [38,39], selective etching [40,41], etc) to switch the adhesion/de-adhesion state of the interface between the donor/transfer medium (e.g. elastomer stamp, fluid) and microLED chips in a highly controlled, scalable, and accurate way. For instance, the micro-transfer printing (µTP) utilized an elastomer stamp with thousands of tiny posts (∼10 800 posts [42]) to control the interface adhesion by tuning the peeling speed. Thus, microscale and even nanoscale structures can be selectively transferred in parallel. X-Celeprint demonstrated that µTP can achieve the transfer reliability (chip size of 8 × 15 µm 2 ) of ∼99% [43], with the maximum transfer efficiency of >6.5 million h −1 [44] and the placement accuracy of ∼1.5 µm [45]. Recently, different kinds of lasers have been introduced to enable digital and parallel transfer processes with highly enhanced throughput, costefficiency, and process flexibility. Especially, extremely high assembly rates (>100 million h −1 ) can be achieved by arrays of laser beams [39,46]. Because of recent significant progress, these mass transfer techniques could become candidates for industrialized manufacturing of microLED displays.
Recently, many review papers have merely focused on microLED displays [5,7,[47][48][49], which mainly concluded the technological path of microLED chips or the solution of full-color displays. Here, we attentively present an in-depth analysis of the latest developments of mass transfer techniques. In section 2, we first discuss the general assembly process (i.e. epitaxial lift-off and pick-and-place process) of microLED displays and highlight key challenges of mass transfer techniques. Then, various state-of-the-art mass transfer strategies and principles adopted in different production steps of microLED displays are described in the subsequent two sections. Finally, we discuss the future opportunities and challenges in this field.

Basic knowledge of mass transfer
The assembly of microLED chips generally includes several key process steps, including releasing massive microLED chips from the donor/growth substrate (i.e. epitaxial lift-off process), adjusting pitch size, and finally aligning and moving them to a backplane/receiver substrate (i.e. pick-andplace process) [3,42]. For achieving full-color microLED displays, mass transfer techniques can be divided into two categories [50,51]. (a) The first strategy is the selective batch  [11]. Reprinted with permission from AAAS. [30] John Wiley and Sons. 'Visible light communication'. [19] John Wiley & Sons. Reprinted by permission from Springer Nature Customer Service Centre GmbH: Springer Nature, Sci. Rep. [31] © 2019. Reproduced from [32]. CC BY 4.0.
transfer of RGB microLED chips, respectively (figure 2(a)) [52]. At the epitaxial lift-off process, microLED chips with different colors are separated from their growth substrates (figures 2(a-ii)). Next, in the pick-and-place process, a transfer medium is used to pick up the RGB microLED chips from different growth substrates (figures 2(a-iii)-(a-v)), followed by a delivery step to place RGB microLED chips to the target receiver substrate (figure 2(a-vi)). (b) The second strategy is the selective batch transfer of monochromatic Blue/UV microLED chips (figures 2(b-ii, iii)), followed by integrating color conversion (figures 2(b-iv, v)) such as ink jet printing of quantum dots (QD [53][54][55]) or light-emitting polymer [56], as shown in figure 2(b). Notably, only for some special kinds of displays with extremely high resolution (e.g. AR [57,58]), the entire blue microLED chips can be directly transferred right after the epitaxial lift-off process (termed as 'monolithic transfer' [59][60][61][62]) with no need for changing the original distance between adjacent microLED chips [63]. However, the color conversion techniques are still plagued by some technical issues, such as difficulties in heat dissipation, low color conversion efficiency, and thermal stability [2,[64][65][66]. Therefore, till now, a mass transfer process is still an indispensable step.
Essentially, a mass transfer technique fundamentally relies on the efficient, reliable, and parallel control of interfacial adhesion at some critical interface, i.e. microLEDs/growth substrate, transfer medium/microLEDs, and microLEDs/receiver. Firstly, the epitaxial lift-off process determines the interface adhesion state of the microLEDs/growth substrate. For the fabrication of microLEDs, a buffer/sacrificial layer must be introduced on the growth substrate (sapphire) to reduce the effect of lattice/thermal mismatch [67]. The GaN buffer layer connects with the sapphire substrate by strong sp3-type covalent bonds (i.e. chemical interactions, ∼1000 kJ mol −1 [41,68]). This strong adhesion makes it very difficult to regulate interfacial adhesion only by a physical stimulus. Therefore, high-energy processes (e.g. laser [69, 70]) or chemical etching is required [40,41]. Due to a complicated coupling between the physical (e.g. laser-induced heating and molten, and shock [39]) and chemical (decomposition of the sacrificial layer [71,72]) processes, mass transfer techniques firstly face a serious challenge of how to control the epitaxial lift-off process selectively and precisely. A wide range of factors, such as the chip thickness [73], structure [74,75], size, and the method for injecting energy (laser or etchant) at the interface for peeling, affect the final interface adhesion state, as well as the quality of the peeled chip. Typically, the epitaxial lift-off process can only enable the direct preparation of a limited range of microLED displays, while for large-area displays with variable pixel arrangements, additional pick-and-place processes are required.
By contrast, physical adhesion (e.g. VDW force [76,77], fluid tension [78][79][80], electrostatic force) plays a critical role in pick-and-place processes. The ability of the transfer medium to switch adhesion from the strong to the weak state in a highly parallel, efficient, and accurate manner is the main factor to determine the capability of mass transfer techniques. During the pick/retrieval step, the adhesion strength of the transfer medium/microLEDs interface must be stronger than that of the microLEDs/donor, thus, allowing for the successful peeling of microLED chips from the donor substrate. During the place/printing step, a low interface adhesion strength is favorable for the transfer of microLEDs to the receiver substrate. Compared with chemical adhesion, some physical adhesion could provide a much lower and variable adhesion strength [68]. For typical reversible adhesive systems, the critical force for the interface separation F c is found to be F c = √ G crit √ A/C [81], where G crit (critical energy release rate) is a property set by the materials comprising the interface, A is the surface area of adhering, and C is the system compliance. This relationship suggests that the physical adhesion of an interface is governed by three key parameters, which are dependent on both the geometry and material properties of the interface [82]. Consequently, how to regulate the above three key parameters by external stimulation to enable reversible adhesive, and how to exert external stimulation in a fast, accurate, selective, and massive way are emphasis and core difficulties of mass transfer techniques.

A brief chronology of the development of mass transfer
Based on different strategies to realize batch control of interfacial adhesion, a series of mass transfer techniques have been developed. A timeline is illustrated in figure 3 to outline the major milestones toward the development of mass transfer. As a prerequisite step of mass transfer, the epitaxial lift-off technique was first developed in 1978 [40,83,84], LEDs with AlAs sacrificial layer were selectively etched by HF, i.e. chemical lift-off (CLO) technique. Soon after, the VDW epitaxial lift-off (VWDE) technique and laser lift-off (LLO) technique was developed in 1997 [85][86][87]. Unlike the CLO technique, the VWDE technique uses a buffer layer of two-dimensional materials to substantially lower the interface adhesion by replacing chemical adhesion with VDW forces [88,89]. To improve the peeling efficiency and quality, lasers were introduced to enable the decomposition of the sacrificial GaN layer. Afterward, LLO became the mainstream [90][91][92]. Meanwhile, the capacity of selective peeling microLED chips provide a simpler solution to adapt the specific pixel arrangement density. Hence, selective lift-off techniques were developed recently [93,94].
Generally, pick-and-place techniques can be classified into three types according to the relative position of the transfer medium and receiving substrate. (a) The contact µTP technique requires microLED chips directly in contact with the receiving substrate with the assistance of a special stamp, which was first introduced by Rogers et al [33,95]. In 2013, LuxVue proposed a new design concept of the contact µTP via the electrostatic force [96,97]. Typical contact µTP techniques can reach a high transfer yield (>99.99%). However, the low transfer rate is a bottleneck. (b) The laser non-contact µTP technique, which introduces a laser to separate the selected chip from the stamp by an interfacial thermal mismatch or blister ejection introduced in 2012 [ [98][99][100][101]. Specifically, these techniques need a certain distance between the stamp and the receiver substrate. The laser non-contact µTP technique can achieve a high transfer rate (∼100 million h −1 ), however, with an unacceptable success rate (∼90%). (3) The self-assembly technique utilizes fluid (e.g. fluid selfassembly in 2008 [102][103][104]) as a transfer medium, and generates gravitational, hydrophobic, or hydrophilic forces to identify and localize microLED chips with specific dimensions. The self-assembly technique could also achieve a high transfer rate (∼99.9%, ∼100 million h −1 [105]), nevertheless, with special requirements for microLED chips and receiver panels.
The continuous progress of mass transfer significantly promotes the display quality and integration scale of microLED displays, as shown in the right part of figure 3. The development of microLED displays can be generally divided into three stages: integration of monochrome micro-displays, highcost integration of large-area displays with low PPI (figure 3. Right), and low-cost integration of any size display with high PPI (future). The microLED display was first invented by Jiang's group in 2000 [2], and the transfer technique at this time could only integrate a few arrays of microLED to form a passive matrix (PM) micro-display. A first 10 × 10 resolution PM microLED display was achieved in 2001 [111], an array of 32 × 32 in 2002 [114], 64 × 64 in 2003, 128 × 96 in 2004, and many more elements were subsequently reported. In 2008, Rogers' group first used the µTP technique to realize a 16 × 16 microLED display [111], which laid the foundation for preparing large-area displays. The first large-area display (55-inch LCD) with 6 million microLEDs launched on the market in 2012, which were fabricated using pick-and-place techniques. In 2014, a full-color microLED display with 1700 PPI resolution was fabricated by contact µTP [115]. To achieve better resolution and brightness, the active matrix(AM)-driving technique had been introduced [38,116]. In 2019, a 3.3-inch full-color AM microLED display (324 µm × 324 µm pixel size) was demonstrated by contact µTP combined with CLO [116]. The most recent large microLED display (∼16 K) with a dimension of 21 m × 5.5 m was presented by Sony in 2019. In recent years, the growing presence of new microLED display products has relied on the progress of mass transfer techniques. However, till now, the high price of manufacturing costs has hindered the popularity of microLED displays.  [86], Copyright (1992), with permission from Elsevier. 'GaN film was peeled by laser lift-off' Reprinted [85], with the permission of AIP Publishing. 'Contact µTP technique by PDMS stamp' [33], Reprinted with permission from [106]. Copyright (2007) American Chemical Society. 'Fluid self-assembly', Reproduced with permission from [107]. 'Blister-based laser-induced forward transfer', Reprinted from [108], Copyright (2013), with permission from Elsevier. 'RTR-µTP technique' [109], John Wiley & Sons. 'Electrostatic µTP by new stamp', From [37]. Reprinted with permission from AAAS. 'Laser contact µTP technique', From [110]. Reprinted with permission from AAAS. 'KSU logo of microLED were integrated' Reprinted from [111], with the permission of AIP Publishing. 'Display was realized by CLO combined with µTP', From [112]. Reprinted with permission from AAAS. 'Large-area display with 55-inch', Reprinted from [113], with the permission of AIP Publishing. 'The Wall' by Samsung", [42]  There is a continuous pursuit of a more cost-effective (relying on extreme high efficiency, yield, and accuracy) mass transfer technique to become the key point of popularity of microLED displays. To facilitate a better understanding of different mass transfer techniques, table 1 summarizes the characteristics, transfer yield, available efficiency, the mechanisms for adjustable interface adhesion, and limitations of all covered techniques. The detailed mechanism of each specific technique will be described in the next section.

Laser lift-off technique
To achieve the cost-efficient mass transfer process, the epitaxial lift-off process with high yield and reliability is the prerequisite. LLO provides a well-established route for the epitaxial lift-off process, as shown in figure 4(a) [38,117,118]. A laser passes through the transparent sapphire substrate (figure 4(a-ii)), which causes a rapid temperature to rise at the interfacial GaN. The high temperature could make the sacrificial GaN layer decompose into metallic gallium and nitrogen (decomposition temperature ∼800 • C [73,119]). As a result, the interfacial adhesion between the sapphire substrate and microLED chips can be largely weakened, and the sapphire substrate can be released by remelting the metal or etching the metal away by HCl (figure 4(a-3)). Various high-power laser sources with photon energy greater than the bandgap of GaN (3.4 eV) have been employed in LLO, such as excimer lasers (248 nm for KrF laser [92,120], 193 nm for ArF laser [121], and 308 nm for XeCl laser [122]) and Q-switched lasers (355 nm laser [119,122] and 1064 nm laser which needs additional InGaAsN sacrificial layers [123]) with a short (nanosecond) pulse width.  [73]. Reprinted from [129], Copyright (2003), with permission from Elsevier. (c) The process window of a reliable LLO process is related to several process parameters. Reprinted from [124], with the permission of AIP Publishing. (d) The distribution of tensile stress of GaN film under Gaussian and uniform laser energy distribution. Reprinted from [125], Copyright 2012, with permission from Elsevier. (e) Schematic diagram of the area-selective LLO process. Reproduced from [130]. CC BY 4.0.
A reliable LLO process should not only achieve complete interface separation but also avoid the occurrence of defects that affects the electronic/optical properties of released chips. The adhesion strength between GaN and the sapphire substrate is closely related to the laser energy density, where there exists a threshold for lift-off. Some thermal/mechanical effects (e.g. cracking in figure 4(b) [73], buckling caused by thermal stress, the vapor pressure of nitrogen, thermal shock, etc) of the high-energy laser could make the damage-free liftoff process become a tough challenge. Numerous studies have revealed that it is important to control the laser processing parameters [124]. Typically, as shown in figure 4(c), the process window of the reliable lift-off is related to laser energy (E P ), spot size (d p ), pulse width (τ ), and GaN thickness (h f ). Also, compared with the laser spot with uniform distribution of energy density, the laser with Gaussian distribution is better to avoid damage, since the latter can cause a smaller temperature distribution at the edges of the laser irradiated zone, resulting in smaller nitrogen pressure and thermal stress (figure 4(d) [125]). To further improve the process quality, several optimizations were proposed. Insertion of a sacrificial layer or blocking layer between GaN and sapphire was a popular choice. For example, the carbon nanotube (CNT) was inserted into the GaN-sapphire interface. The CNT could act as a heating wire to elevate the GaN temperature more effectively to reduce the required energy threshold (from 1.5 to 1.3 J cm −2 ). Another attempt is to introduce the femtosecond laser (350 fs-520 nm laser [69,126]) in consideration of its highly suppressed thermal effect. Especially, the absorption of 520 nm laser relies on two-photon absorption with subbandgap excitation [127].
The conventional LLO is a whole-area peeling process, which must rely on additional assembly techniques to accomplish selective integration (to adapt to required pixel arrangement) of heterogeneous RGB microLED chips into the display panel. To remarkably reduce the difficulties in the following assemble process, selective LLO (SLLO) was developed to enhance the overall efficiency of mass transfer [93,94,128]. In SLLO, a laser is irradiated to selectively separate microLED chips from sapphire. Generally, controlled adhesion between  [67,131] the selected LED and the substrate is key for SLLO. Thus, a lot of studies seek to understand the fundamental relationship between adhesion and laser parameters. Interestingly, figure 4(e) shows a novel SLLO process by a twostep procedure [128]. During the first step (figure 4(e-ii)), the perimeter of the selected area for separation is scanned by a high energy (∼26 J cm −2 ) femtosecond laser for obtaining high-quality boundaries through laser-induced cracking.
In the second step, to avoid chip damage, lower pulse energy (∼9 J cm −2 ) is used to peel the selected region (figure 4(e-iii)).
The experimental results show that this SLLO is capable to fabricate freestanding InGaN/GaN LED chips with thicknesses lower than 5 µm. Till now, SLLO is still in the early stages and its practical effect needs to be further verified.

Chemical lift-off technique
Unlike the LLO technique, microLED chips will not undergo any physical/thermal impact during the CLO process, which removes microLED chips by selective wet etching of sacrificial layer between the device layer and sapphire substrate. Table 2 lists some commonly used sacrificial materials for CLO. A typical choice is Si sacrificial layer. Once the Si sacrificial layer reacts with the KOH solution, it will decompose into K 2 SiO 3 and H 2 , causing the interface separation. Another example is the AlInP sacrificial layer, which can react with HCL to produce reaction byproducts, AlCl 3 , InCl 3 , PH 3 [67,131], and after selective etching and epitaxial release, atomically smooth GaAs surfaces. Meanwhile, to avoid the floating away of microLED chips, protective anchors that will not be etched are made together with microLED chips [132], thus allowing the suspension of microLED chips after etching. And then, once the transfer stamp touches the microLED chips to exert some pressure, the anchor will be broken to fully release to the chip. The relatively low production efficiency is the main problem for CLO since the wet etching of a sacrificial layer often takes several hours [20,143,144]. For example, a 2-inch GaN film with a SiO 2 sacrificial layer needs ∼56 h to achieve 70% peeling area by HF etching [145]. To improve the peeling efficiency, sacrificial layers with nanopores [146,147], nanorods [148], and other special structures (e.g. triangular-shaped hole structure [145], hexagonal-shaped air-void structure [75]) have been developed. Zhang et al introduced GaN nanorods structure by electrochemical etching [74], ∼1 cm 2 freestanding GaN layers (thickness ranging from 500 nm to several microns) can be delaminated in <20 min. Similarly, for the LED structure with triangular-shaped voids at the GaN/sapphire interface, the peeling efficiency and success rate can be increased 2 and 1.3 times, respectively, compared with the original structures. In addition to these special structural designs, the peeling process can be facilitated by introducing additional electric fields or optical fields (i.e. photoelectrochemical/electroCLO technique). UV illumination is used to generate electron-hole pairs on the semiconductor surface, which can enhance the oxidation and reduction reactions in electrochemical cells. Youtsey et al demonstrated that a 4-inch wafer-scale GaN epitaxial material can be peeled off in <2 h. Although significant progress has been achieved in CLO, the peeling efficiency and success rate still need further improvements. In addition, how to achieve selective peeling is still a tough challenge for CLO.

Van der Waals epitaxy lift-off technique
As the latest lift-off technique, the VWDE utilizes 2D materials as the sacrificial layer [89,149,150], such as graphene [1], boron nitride (BN) [151,152], and MoS 2 . There are two significant advantages of this technique. Firstly, 2D materials can overcome the thermal expansion mismatch and inplane lattice constant mismatch (only ∼0.58% [153,154]) between the GaN layer and growth substrate. Secondly, the VDW force between two adjacent layers of 2D materials is much weaker than the chemical interactions of covalent bonds. Therefore, the VWDE can enable a mechanical lift-off process without the assistance of additional chemical solution or laser, as shown in figure 5(a) [87]. Taking the BN sacrificial layer as an example, figure 5(b) shows a detailed process of the interface separation [155]. After applying mechanical loading, the detaching position is always inside the III-nitride epilayer (at the h-BN/h-BN interface), mainly due to the almost freesliding path between hBN/hBN. The energy barrier required to separate the hBN/hBN interface (2.0 meV per atom) is lower than that of the GaN/BN interface (4.5 meV per atom). Therefore, the success of VDWE requires precise control of the layer-to-layer bond strength (i.e. 2D/2D layer) between multilayered structures.  [160], with the permission of AIP Publishing. (c) Photograph, optical microscope, and SEM of GaN/conventional AlN/BN (i), (iii) and GaN/high-temperature AlN/BN (iv)-(vi) grown on sapphire, respectively. Reprinted with permission from [156]. Copyright (2020) American Chemical Society. (d) A theoretical model of the Van der Waals epitaxy lift-off process describes that the peel strength changes with different peeling angles (e) and peeling speeds (f). (e) Reproduced from [157]. © IOP Publishing Ltd. All rights reserved. (f) Reprinted from [161], Copyright (2021), with permission from Elsevier.
To enable a reliable VDWE, it is essential to fabricate a high-quality GaN layer on a 2D-materials-covered sapphire substrate with controllable interface adhesion. For example, a unique approach of controlling the adhesion of h-BN has been reported by Vuong et al, which could achieve both desired lift-off layered h-BN and mechanically inseparable robust h-BN layers [156]. As shown in figure 5(c), the quality of AlN/h-BN structures grown at different temperatures (1100 • C and 1200 • C) is compared through optical and scanning electron microscope (SEM) observations. The conventional GaN/AlN/h-BN structure (1100 • C) enables an easy mechanical lift-off process from the growth substrate by using a transparent tape (figures 5(c-i)-(c-iii)), while the GaN/AlN(high-temperature)/h-BN structure was mechanically inseparable (figures 5(c-iv)-(c-vi)). A possible explanation is that the high-growth temperature leads to Al diffusion into h-BN, which enhances the surface and interface dangling bonds and anchors the layered h-BN. Also, a series of theoretical models (figure 5(d)) indicate that the peeling stress (F), peeling angle (θ), peeling speed (v), and geometric dimension (length and width) of the 2D layer also play important roles in VDWE [157]. Figures 5(e) and (f) present the peel strength change with different peeling angles and peeling speed. The peel strength increases linearly with the peeling speed while decreasing with the peeling angle. Therefore, a large peeling speed with a small peeling angle could be more appropriate for VDWE.
Although substantial progress has been made in VDWE in recent years, there is still a lot of studies that need to be done before mass production. One of the studies is to ensure the film quality of GaN. Especially, the lack of dangling bonds on the surface of 2D materials makes it difficult to ensure nucleation, thus further optimization schemes need to be introduced, such as the introduction of buffer layers (e.g. aluminum buffer layers, whose aluminum is more easily bonded to BN [158]) or nanopatterned sapphire [159]. Meanwhile, area-selective liftoff with large areas is still rarely studied. The need to consider, for example, the effect of inhomogeneities in large-area 2D materials due to local buckling or defects on the VDWE process still needs further study. Finally, further exploration of inexpensive techniques to prepare 2d materials is needed.

Contact µTP technique
As the most common form of pick and place, contact µTP generally utilizes a specially designed stamp as a medium to enable the batch transfer of microchips between the donor substrate and the receiver substrate. At the beginning, a proper preload is applied on the stamp to ensure full contact between the stamp and chips, which should provide enough adhesion to pick up chips from the donor substrate. Then, the stamp is brought into contact with the receiver substrate by precise manipulation of the stamp/chip adhesion to print chips onto the receiver substrate. The whole pick-and-place process is highly demanded in a selective and massively parallel way for high throughput. Thus, the key to successful contact µTP is a timely, flexible and ingenious use of adhesion modulation of the stamp/ink interface. According to the control principle of interfacial adhesion, the contact µTP techniques can be classified as follows: (a) VDW contact µTP technique; (b) Electrostatic/Electromagnetic contact µTP technique.

Van der Waals contact µTP technique.
VDW contact µTP technique utilizes a viscoelastic stamp (e.g. PDMS) to introduce VDW adhesion force. By cleverly manipulating the interface adhesion (denoted as the critical energy release rate G crit for peeling), several advanced transfer printing techniques have been developed. In figure 6(a), Rogers's group verified that the adhesion strength of a viscoelastic stamp is rate-dependent [106,162,163]. The result shows that the adhesion switching ratio (the achievable maximum adhesion strength for high peeling rate divided by the minimum adhesion strength for low peeling rate) is over 3. Based on this adhesion switching ability, as shown in figure 6(b), microLED chips can be separated from the growth substrate with a quick peeling step and printed on the receiver with a much slower peeling process [164]. A higher adhesion switching ratio is directly related to the success rate for contact µTP. RTR contact µTP with a ∼100 adhesion switching ratio was developed by KIMM ( figure 6(c)) [109]. The results revealed that the adhesion of the RTR stamp is not only related to the peeling rate but also determined by the geometry dimension and moving direction of angled posts of the stamp. Figures 6(d) and (e) show the predicted interface adhesion (G) with the rolling speed for different rotation directions and roll radius R. A much higher G can be achieved by forward rolling with a large R. Thus, forward rolling is suitable for separating microLED chips from the growth substrate with an appropriate roll radius. Generally, for the selective batch transfer of microLED chips with specific arrangements and sizes, special PDMS stamps were designed and fabricated with structured micro-pillars. So far, these special PDMS stamps with structured micro-pillars can be fabricated by nanoimprint lithography [165][166][167], laser ablation [168,169], soft lithography [95,170,171], or siliconbased mold fabrication [172,173], when covering a small area. However, it is difficult to maintain the high uniformity of the micro-pillars over large areas. Also, the clustering [174][175][176], self-collapse [177,178], and other phenomena of adjacent micro-pillars during the contact µTP process need to be strictly avoided, which still causes many difficulties.
To enable the substantial control of the interface adhesion for an enhanced transfer success rate, several improvements of VDW contact µTP have been proposed by modifying the contact area or elastic modulus of the stamp, such as the surface-relief-structured stamp with a ∼1000 adhesion switching ratio [95], and an inflatable stamp with a ∼50 adhesion switching ratio [181], etc. These improvements have been well summarized in some other reviews [77,182,183,190]. In addition to the improvement of the adhesion switching ratio, another attempt is to enhance the efficiency and flexibility of the transfer process by incorporating lasers in contact µTP, namely laser-assisted contact µTP [39,98,184,186]. The selective transfer can be easily obtained by controlling the shape of the laser spot, as well as the scanning paths, which can avoid the use of the above-mentioned sophisticated stamps with microstructures. Figure 7 shows several representative laser-assisted contact µTP techniques. Huang et al have demonstrated that reversible VDW adhesion can be achieved by utilizing the shape memory (SMP) stamp with pyramid posts, as schematically described in figure 7(a) [184]. At first, an SMP stamp with pyramid posts is heated and pressed onto the target chip for achieving a large contact area, then its shape is fixed by cooling ( figure 7(a-i)). Subsequently, the chip was peeled from the growth substrate due to the high adhesion ( figure 7(a-ii)). For the printing step ( figure 7(a-iii)), a laser heats the SMP to initiate shape recovery to largely reduce the contact area,   [110]. Reprinted with permission from AAAS. (d) Schematic illustration of the gripping and release process of the shape memory gripper (right). Experimental results of the functional relation between and grip strength F grip−force and grip speed at temperatures of 120 • C and 30 • C. From [189]. Reprinted with permission from AAAS. thus, the chip could be successfully transferred ( figure 7(a-iv)). Theoretical analyses ( figure 7(a), right) show that the regions of successful transfer printing are determined by the dimensionless constant W 1 / l and γl/ Eh 0 , where W 1 , l, h 0 are the spacing, height, and length of the trapezoidal posts, E is the elastic modulus of the stamp, and γ is the work of adhesion [189]. Therefore, a rational design of the pyramidal post is required. The pyramid posts could be further replaced by vertical posts (figure 7(b), left [185]). The lateral displacement of a vertical post was induced by laser heating, which causes a mixed-mode loading at the interface for decreasing adhesion ( figure 7(b-iii)). Similarly, a fracture mechanics model is developed to calculate the pull-off force (in figure 7(b), right). The pull-off force F decreases with the shear strain ε strain , and it can be predicted by the scaling law: F where Γ 0 is fracture toughness of the stamp/micro-device interface [187], E stamp and L post are the stamp elastic modulus, post width, and post height, respectively. More recently, Wang et al invented a simple shapeconformal stamp with micro-spheres [110]. In the pick-up process, due to the strong viscosity of the shape-conformal stamp, the device is peeled off from the donor substrate. In the printing process, the extremely large volume expansion of microspheres (figure 7(c-ii)) can rapidly induce hierarchical microstructures on the surface of the adhesive layer and decrease the energy release rate markedly by heating ( figure 7(c), right), and the adhesion can be manipulated in the range of ∼100 times of magnitude. To massively assemble the objects with a wide range of scales, shapes, and quantities, a universal SMP block has been invented (figure 7(d) [188]). Since the stiffness of the SMP can reversibly change ∼1000 times under laser stimulus, objects can be embedded in the SMP under pressure and grasped ( figure 7(d-i)). Then, the shape recovery upon the stimulation of a laser could facilitate the releasing process ( figure 7(d-iii)). The adhesion could be reduced in three orders of magnitude (figure 7(d). Right).
Despite these achievements in laser-assisted contact µTP, there are still some bottlenecks that need to be addressed. Typically, the laser-induced adhesion switch mostly is a thermal activation process, however, the absorption of the laser still relies on the surface of the chip. The challenge is how to effectively avoid possible thermal damage on chips. An improved method is to add some laser absorbing material on the stamp surface, such as carbon black particles for absorbing 808 nm laser [191,192]. Moreover, little research has investigated the actual transfer efficiency and reliability of laser-assisted contact µTP techniques.

Electrostatic/electromagnetic contact µTP technique.
Electrostatic/magnetic contact µTP techniques utilize some special stamps that can generate switchable electric/magnetic force. Representatively, as shown in figure 8(a) [37], a groundbreaking electrostatic stamp was developed by Hart et al, which is composed of soft nanocomposite electroadhesive (SNE, multi-walled Al 2 O 3 -coated CNTs, diameter of ∼20 nm) with a conductive bottom electrode (titanium nitride, TiN). When an appropriate voltage (>100 V) is applied, the strong electrostatic adhesion force of the SNE stamp can overcome the adhesion between the target chip and donor substrate ( figure 8(a-ii)) for picking up. Since the intrinsic dry adhesion of SNE is very low, turning off the voltage allows the easy release of the object at the desired location ( figure 8(a-iii)). As shown in figures 8(b) and (c), the external voltage, CNT fiber density, and Al 2 O 3 thickness are the major factors that affect the electrostatic adhesion of the SNE stamp. The adhesion switching ratio increases with the external voltage, and CNT density decreases with the Al 2 O 3 thickness. Therefore, with a proper design of the SNE stamp, this electrostatic contact µTP technique could pick up objects with sizes of ∼20 nm to 100 µm.
Compared to the electrostatic contact µTP technique, the electromagnetic contact µTP technique requires the microLED chips with an additional magnetically sensitive layer to generate magnetic force [193,194]. The electrostatic/electromagnetic contact µTP technique envisions a promising approach for the integration of extremely small objects. However, the mass transfer of tiny chips with predefined patterns remains a significant challenge, which fundamentally relies on the preparation and patterning of the stamp. As shown in figures 8(d) and (e), the above-mentioned SNE stamps (∼100 µm long) can be patterned by CVD, chemical vapor deposition methods [195][196][197]. However, the large-area preparation of CNT arrays is still in a preliminary stage, with typically reported preparation abilities of only ∼200 g h −1 and an area of ∼300 mm 2 . Meanwhile, the cyclical stability, the influence of the charge accumulation, and the homogeneity of magnetic material (for the electromagnetic contact µTP) require further verification.

Laser non-contact µTP technique
For a non-contact µTP process, the stamp does not directly contact the receiver substrate. Thus, the success of the placing process is independent of the topography and properties of the receiving surface. Without the assistance of the receiver substrate, the stamp should achieve infinite adhesion switchability to launch the selected chip away from the stamp. The laser-driven non-contact µTP is the only technique that can be manipulated in a non-contact transfer mode reported so far, resulting from the remotely injected energy (laser) at the chip/stamp interface. Generally, for a laser non-contact µTP process, a shaped laser irradiates through a transparent substrate, bringing about physical changes or chemical reactions at the interface, such as local deformation, vaporization, and thermal stress mismatch, to achieve the switch-off of the chip/stamp adhesion. Therefore, a selective transfer process with local, non-contact outer stimuli (laser) could greatly enhance the versatility of mass transfer techniques. The most significant advantage of laser non-contact µTP is the scalable chip size, shape, transfer frequency, and selective/parallel mode, which are mainly determined by the laser beam that can be flexibly controlled by the setup of laser equipment and optics. In this section, three types of laser non-contact µTP techniques are reviewed.

Laser-induced forward transfer (LIFT).
LIFT is a major kind of laser-assisted printing technique allowing the deposition of a wide range of functional materials/structures into user-defined patterns. Figure 9(a) illustrates a typical LIFT process including five main components [198][199][200][201]: a laser pulse, a stamp with a transparent substrate, and a laser absorbing layer (dynamic release layer, DRL), a target chip, and a receiving substrate [199,202]. A laser passes through the transparent substrate and is absorbed by the DRL, leading to a fast increase in the interfacial pressure due to the ablation or vaporization of the DRL. The high pressure will push the chip from the donor to the receiver. The common choices of DRLs are triazene polymer (TP) and metal films (Ti, Au, and Pt) with a thickness of hundreds of nanometers. In particular, TP is a kind of polymer that contains aryl-triazene chromophores with high absorption at UV wavelength. It can be decomposed at very low laser fluence (only ∼50 mJ cm −2 [203,204]), thus making it easier to enable the transfer process and reduce the risk of thermal damage.
However, there are still some considerable performance bottlenecks of LIFT that will bring tough difficulties to mass transfer. One challenge is the generation of shock waves during the LIFT process. As shown in figure 9(b), by timeresolved shadowgraph, the shock wave is found to be faster than the chip, which will be reflected by the receiver and interact with the flying chip [205,206]. This interaction could cause unwanted deflection of the chip orientation, as well as chip damage. Optimization of some key process parameters, such as the laser energy, the gap between receiving substrate and stamp, the DRL thickness, and environmental pressure has been tried to avoid the impact of shock waves. Shaw-Stewart et al revealed that the intensity of the shock wave decreases with the gap distance (<0.5 mm) at normal atmospheric pressure, so it is possible to eliminate the shock wave by choosing a larger gap (∼1 mm [207]). Also, the environmental pressure can be regulated, such as less than 100 mbar or in a vacuum, to enable a safe transfer in a gap distance of less than 20 µm.
Another challenge is how to achieve the non-destructive transfer of ultrathin chips. Even in a vacuum, although there is no air resistance and shock waves, it is still difficult to ensure the safety of the chip due to the extremely high impact velocity (∼1200 m s −1 [206]). Lower laser energy certainly can result in a smaller impact speed [206], thus, some specific strategies have been developed, including reducing the DRL thickness and introducing smart intelligent spot systems [210]. Rapp et al proposed an intelligent beam shaping (SBS), as shown in figure 9(c-ii) [210]. Different from a conventional LIFT beam system (CB, line 1, figure 9(c-i)), SBS requires another mask system, i.e. a second line (Line 2) on Line 1. Line 2 is created by using a beam splitter before Mask 1 so that the laser beam passes through the second mask (Mask 2). In addition, an additional energy attenuator is placed after Mask 1 (Line 1) to reduce the energy in the center of the spot. The resulting irradiation on the stamp is the combination of Masks 1 and 2, as presented in figure 9(d-iii). This setup allows for higher energy at the edge of the spot (figure 9(d-ii)) while maintaining a lower and more uniform energy distribution in the central area ( figure 9(d-i)). Due to this specially designed energy distribution, only the laser energy at the edge (Line 2) is required to reach the transfer threshold, while allowing the center energy (Line 1) to be reduced to half of the transfer threshold, thus reducing damage. Although lots of achievements were made in LIFT, it still needs to be further improved to form a stable process window.

Blister-based laser-induced forward transfer technique.
To fundamentally avoid the impact of shock waves and minimize the risk of chip damage, a blister-based laserinduced forward transfer technique (BB-LIFT) was developed [211][212][213]. Unlike the traditional LIFT, BB-LIFT utilizes a double-layered stamp that is composed of a DRL (e.g. 15 µm thick polyimide [100,214] or 80 µm thick aluminum [215,216]) and an adhesive layer (e.g. 70 µm thick PDMS [216,217]) to bond chips. During laser irradiation, only a small portion of the DRL is ablated to provide impact energy via gas products. The DRL can encapsulate the shock wave inside by creating an expanding blister that serves as a soft ejector to push the chip more gently toward the receiving substrate, as shown in figure 10(a) [108]. Thus, instead of directly using the impact energy of the ablated DRL, the created blister can also realize a quick release (the transfer process takes <50 µs) with higher accuracy (reported minimum placement error of 1.8 µm) and less damage. Moreover, the most significant advantage of BB-LIFT is high scalability. Parallel laser-enabled transfer (MPLET) technology was developed to achieve an extremely high assembly rate (>100 M units h −1 [46]) by simultaneously transferring a large number of chips just by one laser pulse. As shown in figure 10(b), an array of laser beams irradiates multiple chips together, which are divided from a single laser beam with the help of a diffractive optical element. Since each beam can transfer one chip, the maximum assembly rate can be multiplied by the number of diffracted beams.
The same as conventional LIFT processes, the high falling speed (e.g. ∼540 m s −1 for 10 ns-1064 nm laser with aluminum as DRL [215,216]) still becomes a tough problem, which brings difficulties for controlling the transfer accuracy. A statistical test indicated that the transfer error, which mainly includes the lateral offset and the angular deflection, is positively related to the gap between the stamp and the receiver substrate. Unfortunately, when setting the gap very close, the squeeze-film effect may reduce the kinetic energy of the released chip, which will also affect the transfer accuracy [100]. Therefore, it is difficult to accurately control the placement accuracy of multi-beam lasers. Under practical conditions, affected by many factors such as laser energy fluctuations, inconsistency of multi-beam size/shape, initial alignment deviation, and the heterogeneity of material/geometric properties, etc, some unexpected results might occur after laser irradiation, including the failure of chip peeling, unacceptable deviation, and chip damage (creaking), etc. The industry usually uses the experimental trial-and-error method to get a better combination of process parameters. However, up to now, MPLET can achieve a success rate of only about 90% [46], which is far below the desired value. Due to the lack of in-depth research on transfer mechanisms, it is still unable to fundamentally reveal the reasons for the transfer failure, as well as the influence of process parameters on transfer results. Therefore, it still requires a lot of effort in experimental and theoretical research to gradually improve the process stability and reliability.
Another limitation of BB-LIFT is that the stamp cannot be reused due to the irreversible ablation of DRL. As shown in figure 11(c), for the cost-benefit consideration, a reusable BB-LIFT technique was developed [46] based on an ingenious design of a reusable stamp, which is composed of a microcavity with a metal layer attached to the cavity wall and an elastic adhesive stamp with microstructures for encapsulating the microcavity and bonding the chip. Under the irradiation of an 808 nm laser (∼400 mW), the metal layer absorbs the laser to generate heat, which causes the expansion of the internal air of the cavity due to the rapid temperature rise (∼100 • C in 100 ms). The deformed stamp could largely reduce its adhesion (adhesion switching ratio >1000). Figure 10(d) shows the ability to print a microLED chip (400 × 200 × 90 µm) on a variety of tack-free flat or curved surfaces, such as a PDMS column array with a column diameter of 100 µm and a column pitch of 150 µm ( figure 10(d-I)), a steel ball with a diameter of 1 mm (figure 10(d-II)), and a sheet of paper (figure 10(d-III)). Figure 10(d-iv) shows microLED chips were transferred onto the curvilinear surface of a Scindapsus leaf to form a letter 'Z'. The same square microLED chips were printed on a mobile-phone shell with a pit to form a letter 'P', as shown in figure 10(d-V). These presentations demonstrated the powerful ability to accurately place microLED chips regardless of the receiver surface properties. Nevertheless, how to further reduce the geometric sizes of this cleverly designed stamp with high manufacturing accuracy is a big problem for this technique.

Laser-induced forward transfer based on the thermal mismatch.
A new LIFT-type technique based on the thermal mismatch of a multi-layer structure (LIFT-TM) combines the accuracy and versatility of traditional transfer printing processes in picking up chips directly and selectively from their growth/fabrication substrates and has the advantage of conventional LIFT processes to realize noncontact placing process [98,[218][219][220]. As shown in figure 11, the LIFT-TM utilizes an elastomeric stamp made of PDMS to selectively pick up chips to be transferred. Then, a NIR pulsed laser is focused on the stamp-chip interface and the chip absorbs the laser and generates heat, which makes the chip released from the stamp resulting from different thermomechanical responses of the chip and the PMDS stamp [98]. An accurate thermomechanical model has been developed to establish a scaling law governing the delamination of a chip, as shown in figure 11(b). The energy release rate at the stamp/chip interface is characterized by t delamination , which is a function of total heat flux q total /q 0 and the normalized width of the silicon chip (L ′ si = c PDMS ρ PDMS L chip / (c chip ρ chip h chip )), where t delamination and q total are the delamination time and total heat flux, L chip is the width of the chip, c i (i = PDMS, chip) is the specific heat of stamp and chip, ρ i (i = PDMS, chip) is the mass density of stamp and chip, t 0 , q 0 are the normalized characteristic time, normalized heat flux, and normalized width determined by properties of the stamp, chip and their interfaces in the system, respectively. Thus, theoretical models can serve as powerful tools for determining the critical process parameters, estimating the temperature, and calculating stresses at the interface.
Extensive viability tests were conducted by using a laser with an 805-nm wavelength, which demonstrated the wide compatibility of various surface characteristics of the receiver and the ability to place ultrathin microstructures [99]. Figure 11(c-i) shows examples of an array of silicon chips that were transferred onto a silicon substrate to bridge gold traces. Multiple silicon chips (100 µm × 100 µm × 3 µm) were successfully printed layer by layer with high precision for the construction of 3D assemblies in figure 11(c-ii). To demonstrate the ability to transfer chips to non-flat substrates, a 320-nm-thick silicon chip was printed on a ceramic microsphere ( figure 11(c-iii)). Moreover, the microLED chip is still functional after being placed on a silicon substrate ( figure 11(c-iv)). Notwithstanding all this, a foreseeable challenge of this process is the limitation to chip materials. Chip materials with high reflectivity or low absorptive to the NIR laser are incompatible. Also, the required high temperature (100 • C-200 • C) at the interface may bring damage to the chips. Also, the shear forces caused by the thermal mismatch cause a relatively long delamination time (∼0.5 ms), which may limit the maximum transfer efficiency. Moreover, much research still needs to be done to quantitatively discuss important process parameters and their impact on the process accuracy. Figure 11(d) shows the maximum lateral error after transfer versus the receiving substrate spacing (i.e. standoff height [99]). The transfer error can be negligible when the standoff height is less than 20 µm. However, other factors have not been carried out yet.

Self-assembly techniques
Self-assembly is a process that a massive number of independent components can spontaneously form into ordered arrangements. The traditional mass assembly techniques generally include four steps, i.e. positioning the donor substrate, picking up, positioning the receiver substrate, and placing. Each of these steps has become extraordinarily difficult when facing the massive number of chips with microscale sizes. A significant difference from the working principle of the aforementioned techniques is that the self-assembly technique requires only one step, i.e. placing the microLED chips in the corresponding fluid medium, which enables the microLED chips to be oriented and positioned in a predetermined location through medium transport and shape recognition. As an ingenious approach, self-assembly techniques use several physical force fields that can orient individual chips and filter out all parts that are in the wrong positions. These abilities may well greatly simplify the whole working procedures to largely increase the transfer efficiency. According to the specific physical fields for transferring and orientation, self-assembly techniques can be divided into two categories, i.e. fluid self-assembly, and magnetic/electric self-assembly.

Fluid self-assembly.
Representative fluid selfassembly is illustrated in figure 12(a). A specially designed receiver substrate is placed in a self-assembly bath at a specific angle such as ∼30 • to help the chips move gently due to gravity [221]. Once freestanding microLED chips are submerged in the liquid medium (e.g. acidic ethylene glycol), the liquid medium allows microLED chips to move freely until they come into touch with the molten alloy at the predetermined locations to form a stable bonding. microLEDs are positioned and oriented by shape recognition and surface tension-driven self-alignment. Shape recognition is achieved through complementary shapes (pits) of microLEDs on the receiver substrate [222,223]. Thus, as shown in figure 12(b), only microLEDs with specific shapes, such as a round shape with a 160 µm radius, can be adaptable for shape recognition. In contrast, surface tension-driven self-alignment uses capillary forces of the molten alloy on the substrate (figure 12(c) [224]). At first, a confined liquid drop (molten alloy) is arranged in a pre-designed pattern (figure 12(c-i)). The molten alloy spontaneously forms a meniscus shape due to capillary forces (figures 12(c-ii, iii)). This action tends to align the chip to the target location ( figure 12(c-iv)). Generally, an essential design criterion for capillary self-alignment is that the liquid molten alloy constrained by a receptor and a chip (or a part of it) possesses sufficient high potential energy to recover to its equilibrium shape so that this recovery process can sustain the self-aligning motion. Their strong abilities of the orientation of microLEDs have been well demonstrated. The results indicate fluid self-assembly can achieve a 97% transfer rate with self-assembly rate of 40 000 units h −1 for 100 µm-sized components if only mechanical positioning (without electrical connections) is considered [107].
To further promote transfer efficiency, Jacobs et al introduced an automated RTR fluid self-assembly system, as illustrated in figure 12(d). This system contains two parts [25,80,102]: (a) RTR assembly part (shaded in gray), which consists of motor, rollers, customized agitator, and polyimide web with precisely controlled process parameters such as web moving speed and tension. (b) The component recovery and dispensing unit (shaded in blue), primarily utilizes the principle of a jet pump to transport un-assembled chips upward in a closed fluid channel, releasing them above the substrate and tumbling them down under gravity. It has also been revealed that the RTR fluid self-assembly process requires strict control of gravity (i.e. the tilt angle of the receiving substrate) and vibration frequency (0-30 Hz). The effect of the web angle on the assemble rate is shown in figure 12(e). A slight decrease in assembly rate can be observed at higher tilt angles, mainly because the chips start sliding too fast, and some of them slide over the solder bumps, instead of being captured immediately. The assembly rate as a function of vibrational frequency is shown in figure 12(d), the optimal frequency should be >20 Hz. Otherwise, at a too low vibration frequency, the agitation is not strong enough to overcome non-specific adhesion and gravity, and most chips remain stationary. Through a series of optimizations, RTR fluid self-assembly can achieve a assemble efficiency of 15 000 chips (square 500 µm chips) per hour with a yield rate of >99%.
Looking ahead, a lot of further work still needs to be done on fluid self-assembly techniques to meet the requirements of mass transfer techniques. On the one hand, most of the existing fluidic self-assembly techniques stay on some simple shapes of microLED chips (e.g. above-mentioned round and square). Therefore, to improve the universality, more complex 3D geometries of microLED chips need to be investigated [225]. On the other hand, further optimization of chip size, solder materials and assembly automation systems is necessary. Typically, to address interconnect stability and improve transfer yields, solder stacks with low melting point shells (e.g. Bi 33.7 In 66.3 solder shell [222,226]) can be further introduced so that after assembly at low temperatures, a short reflow step at a higher temperature can be used to form the alloy to increase the melting point of the interconnect.

Magnetic/electric fluid self-assembly.
A stronger self-alignment ability is the most salient feature of the magnetic/electric assembly techniques. For the magnetic self-assembly, a magnetic field between the microLEDs and the receptor site is required. As shown in figures 13(a), a thin strip of ferromagnetic material (e.g. nickel, iron, etc) is deposited at the microLED surface as a p-type contact and the adjacent metal pad acts as an n-type contact [227]. For permanently magnetized hard magnetic materials, self-assembly may not require an external magnetic field. By contrast, the magnetization of soft magnetic materials is highly dependent on the external magnetic field and they tend to revert to a non-magnetized state with the removal of the magnetic field. Thus, magnetic self-assembly of soft or superparamagnetic components requires an external magnetic field. During the magnetic assembly process, the microLED chips are placed on a vibrating magnetic substrate ( figure 13(b)). Magnets with opposite polarity are arranged adjacent to each other on the table. The center of the edge between the adjacent magnets becomes the node of maximum magnetic field strength. Once the magnetic microLED chips are placed on the vibrating magnetic table, they can be randomly dispersed by vibration. With the assistance of the magnetic field, the chips are arranged parallel to each other at the nodes of maximum magnetic field strength to achieve the self-assembly process.
In electric self-assembly, a phenomenon of electrophoretic transport (charged particles move toward the electrode opposite to their electric polarity) is applied, which makes the chips move and locate at the selected receptor sites. As shown in figure 13(c-i), O'Riordan et al recently reported a preliminary idea of the programmable electric self-assembly (i.e. field configured assembly, FCA [228,229]), where the electric fields can be configured by selective addressing receptor sites pre-patterned onto a silicon substrate. The silicon substrate consists of 4 × 4 circular arrays of receptor electrodes (100 µm diameter and 250 µm spacing) with counter electrodes (500 µm diameter) located at the four corners of each array ( figure 13(c-ii)). The conductive interconnect tracks connect each electrode to a unique contact pad located at the periphery of the chip (figures 13(c-ii, iii)). The electric field simulation in figure 13(d-I) shows that an electric field can indeed be generated between a unique receptor electrode point on the silicon chip and the counter electrodes at the four corners with a magnitude sufficient to affect electric-field-assisted device transmission. Figures 13(D-ii)-(D-iv) shows a successful transmission of a microLED chip with a 50 µm diameter. Initially, a microLED chip was located at the top right receptor electrode, and after applying an electrical bias of approximately 20 V to each adjacent target receptor electrode, in turn, the chip was delivered to the bottom right receptor finally. It is worth noting that it is still necessary to ensure the shape of the microLEDs is commensurate with the receptor.
Due to the extremely high requirement on the success rate for mass transfer, until now, the above self-assembly technologies cannot be applied in large-scale production, therefore, transfer yields related data are still available. There are still many technical problems to be solved to avoid the missed chip as little as possible. Thus, deep studies of the effect of process parameters on alignment accuracy are needed, such as the strength of the magnetic/electrostatic field, the distance and direction between the chip and site, the size and shape of the chips, and the degree of magnetization/electrostatics. Another limitation is the microLED chips for self-assembly need special structural characteristics and functional properties. For example, magnetic self-assembly can only be applied with microLED chips with magnetic components, while electric fluid self-assembly requires charged microLED chips.
Also how to transfer different colors of RGB microLED chips separately is currently unclear.

Conclusion and future perspective
Mass transfer technique offers the possibility of large-scale, highly efficient and high-yield heterogeneous integration of microLED arrays into spatially organized, functionally arranged 2D and 3D layouts. It creates a number of new applications and industrialization opportunities for next-generation microLED displays and new microLED-based electronics, which are impossible by conventional pick-and-place techniques. In the past decade, the development of mass transfer is generally converging on realizing the following targets: (a) satisfy the needs of cutting-edge applications that hinder from the minuscule size of microLEDs and solve the relevant issues including high accuracy positioning, and selective integration with large numbers; (b) replace the complex, highly expensive, and timeconsuming multiple steps of conventional assembly methods with digital/parallel processes to greatly enhance the throughput, cost-efficiency, and manufacturing flexibility; (c) broaden the process window that can enable a deterministic and reliable transfer process and tolerate a little more process variation due to the hardly avoidable fabrication error for high-reliability requirements.
Recently, various techniques based on ingenious design concepts and processing principles have been established and developed toward the above targets. All these techniques are based on clever manipulation of the interfacial adhesion between the microLED chips and the growth/intermediate substrate. In this review, recent advancements in mass transfer techniques were overviewed. The necessary considerations to properly utilize interfacial adhesion control for desired purposes were summarized. Two essential techniques, which are epitaxial lift-off techniques and pick-and-place techniques were introduced from basic mechanism to the state-of-the-art demonstrations. Despite recent progress, there is a continuing need for further development before these mass transfer techniques become reliable manufacturing candidates for the industrialized manufacturing of microLED displays.
Firstly, a further in-depth study of interfacial adhesion mechanisms is necessary. The essential relationship between the process parameters and interface reaction (e.g. adhesion strength, fracture mechanics, and chip peeling/flighting state) can undoubtedly provide quantitative guidance for the high reliability of mass transfer. For example, as in the case of epitaxial lift-off techniques, a complicated film delamination mechanism involves the synergistic effects of applied physical fields (e.g. mechanical loads and laser irradiations) and chemical reactions (e.g. ablation in LLO, chemical etching in CLO), thus building complex coupling models of physics, mechanics, and chemistry is essential. As another example, in fluid self-assembly, the interface surface energy is an important parameter needed to be carefully considered, however, there still lacks a direct measurement method at a microscopic scale, which may increase the uncertainty of this technique resulting from some indeterminate process parameters (can only be determined by trial and error). The continuing efforts to develop more simple and accurate methods for interface bonding energy measurement and establish corresponding theoretical models are still necessary, which will facilitate a more deep understanding of some interface microscopic phenomena.
Secondly, although lots of small-area transfer schemes have been extensively verified, a further improvement in the reliability, accuracy, and efficiency of large-area transfer techniques is still an area of tremendous opportunities. For contact µTP techniques, the parallel transfer can only be realized by expanding the number of transfer heads. However, the microcolumn of an elastic stamp tends to collapse during contact transfer, and it is difficult to make individual transfer heads in a small size for an electromagnetic or electrostatic stamp. These challenges can probably be overcome by further optimization of the structural design and manufacturing processes of transfer heads. For laser non-contact transfer techniques, although the time for a single transfer is considerably short, the transfer rate is still greatly limited by the moving speed of the laser spot and the number of chips that can be transferred by a single spot. Inncreasing the moving speed of the spot by vibrating mirror scanning (the high-speed vibrating mirror can achieve tens of thousands of scans per second) and the number of spot arrays by optical projection mask and optical space modulation is the future trend. As for the magnetic/electric fluid self-assembly, arrayed magnetic and electric fields need to be introduced. And for contact µTP techniques, the parallel transfer can only be realized by expanding the number of transfer heads. However, the micro-column of an elastic stamp tends to collapse during contact transfer, and it is difficult to make individual transfer heads in a small size for an electromagnetic or electrostatic stamp.
For achieving high reliability, the development of high precision micro-fabrication/position technologies and material preparations for chips, transfer stamps, and receiver substrates/solders are hotspots, since lots of mass transfer techniques rely on some specific properties/shapes at specified positions. For example, the fluid self-assembly technique has special requirements for adhesives, which ask for better fluidity, high bonding selectivity between assembled and unassembled surfaces, and acid/high-temperature resistance considering subsequent processes. However, the present acrylate heat-curing adhesives and solders do not fully meet these requirements. In addition, to improve the transfer accuracy, the introduction of optical position calibration systems or auxiliary positioning technologies may become a focal point. Until now, the optical alignment for contact µTP is still challenging, and further improving sensor resolution and marking stamp materials with better contrast is necessary to achieve a brighter fiducial mark. On the other hand, Emine Eda Kuran et al have proved that LIFT can combine with a magnetic-assisted alignment technology. The chip was transferred by LIFT firstly, while the autonomous positioning can be achieved with the help of magnetic traction.
Further exploration of cost-efficient mass transfer techniques is full of challenges as well as opportunities. The development of reversible laser-assisted µTP has important practical significance which can easily achieve high throughput for large-scale and high-output manufacturing with the help of parallel laser systems and automated platforms. To reduce fabrication costs, a reusable stamp is essential. However, the laser-induced physical/chemical modifications of the reusable stamp need to be carefully controlled to avoid undesirable distortion or damage. In addition, most of these stamps rely on phase transition of smart materials (e.g. SMP), therefore how to further reduce response time through technical innovation of materials and configurations is worth investigating. Also, inventions that can simplify processing steps to reduce the manufacturing cost are favorable. For example, the development of selective epitaxial lift-off processes (specific microLEDs are directly transferred from the source substrate to the target substrate) may probably avoid a series of intermediate steps. Moreover, decreasing the cost of mass transfer equipment deserves the same consideration. The commonly used excimer laser systems (cost millions of dollars) may be replaced by relatively low-cost laser systems (solid-state laser or fiber lasers).
Finally, looking into the future, with a series of tremendous efforts, mass transfer techniques can prepare microLEDs displays of various sizes at low cost. As a result, OLED and LED displays will be replaced, as microLEDs have more advantages for OLEDs and LCDs. The application of microLEDs is not limited to displays, but also will be combined with optical communication, wearable, smart lights and other applications. Especially, high-end glass-based AR/VR can provide a high brightness of 100 000 nits, high ppi > 2000, and high contrast of 100 000:1 to view in an outdoor environment. This brightness cannot be provided by OLED. Since wearable micro-LEDs can be prepared in large areas, wearable micro-LEDs can be widely used not only for biomedical applications (including biosensors, medical devices and optogenetic stimulators) but also for large area detection in devices such as aircraft and vehicles. Also, higher brightness than broadarea OLED at high current density is a significant advantage of microLEDs in the application of VLC, as one of the disadvantages of VLC is the short communication distance. Therefore, in the future, integrated microLEDs can achieve higher data transmission (e.g. 12Gb/s [230,231]), leading to an accelerated convergence of communication techniques interactions.
In summary, the excellent performance of will certainly revolutionize the applications of displays soon. A lot of potentially disruptive technologies for mass production of microLED displays may be developed by a further in-depth study of laser-induced interface phenomena, multiscale adhesion-controlled interface, micro/nanomanufacturing technology, etc. These considerations suggest that this field of study will remain active and dynamic, promising more versatile, effective, and cost-efficient mass transfer techniques for microLED displays fabrication in the near future.