Inkjet printing technology for increasing the I/O density of 3D TSV interposers

Abstract

Interposers with through-silicon vias (TSVs) play a key role in the three-dimensional integration and packaging of integrated circuits and microelectromechanical systems. In the current practice of fabricating interposers, solder balls are placed next to the vias; however, this approach requires a large foot print for the input/output (I/O) connections. Therefore, in this study, we investigate the possibility of placing the solder balls directly on top of the vias, thereby enabling a smaller pitch between the solder balls and an increased density of the I/O connections. To reach this goal, inkjet printing (that is, piezo and super inkjet) was used to successfully fill and planarize hollow metal TSVs with a dielectric polymer. The under bump metallization (UBM) pads were also successfully printed with inkjet technology on top of the polymer-filled vias, using either Ag or Au inks. The reliability of the TSV interposers was investigated by a temperature cycling stress test (-40 °C to +125 °C). The stress test showed no impact on DC resistance of the TSVs; however, shrinkage and delamination of the polymer was observed, along with some micro-cracks in the UBM pads. For proof of concept, SnAgCu-based solder balls were jetted on the UBM pads.

OPEN

Microsystems & Nanoengineering (2017) 3, 17002; doi:http://dx.doi.org/10.1038/micronano.2017.2

Web End =10.1038/micronano.2017.2

http://www.nature.com/micronano

Web End =www.nature.com/micronano

ARTICLE

Inkjet printing technology for increasing the I/O density of 3D TSV interposers

Behnam Khorramdel1, Jessica Liljeholm2,3, Mika-Matti Laurila1, Toni Lammi1, Gustaf Mrtensson4,5, Thorbjrn Ebefors2, Frank Niklaus3 and Matti Mntysalo1

Interposers with through-silicon vias (TSVs) play a key role in the three-dimensional integration and packaging of integrated circuits and microelectromechanical systems. In the current practice of fabricating interposers, solder balls are placed next to the vias; however, this approach requires a large foot print for the input/output (I/O) connections. Therefore, in this study, we investigate the possibility of placing the solder balls directly on top of the vias, thereby enabling a smaller pitch between the solder balls and an increased density of the I/O connections. To reach this goal, inkjet printing (that is, piezo and super inkjet) was used to successfully ll and planarize hollow metal TSVs with a dielectric polymer. The under bump metallization (UBM) pads were also successfully printed with inkjet technology on top of the polymer-lled vias, using either Ag or Au inks. The reliability of the TSV interposers was investigated by a temperature cycling stress test (40 C to +125 C). The stress test showed no impact on DC resistance of the TSVs; however, shrinkage and delamination of the polymer was observed, along with some micro-cracks in the UBM pads. For proof of concept, SnAgCu-based solder balls were jetted on the UBM pads.

Keywords: heterogeneous three-dimensional (3D) integration; inkjet printing; interposer; microelectromechanical system (MEMS);

reliability; super inkjet (SIJ); through silicon via (TSV)

Microsystems & Nanoengineering (2017) 3, 17002; doi:http://dx.doi.org/10.1038/micronano.2017.2

Web End =10.1038/micronano.2017.2 ; Published online: 10 April 2017

INTRODUCTIONOne of the major technology drivers in microelectromechanical systems (MEMS) is to incorporate increasing numbers of functionalities within the same foot print. This increase in functionalities directs the focus toward miniaturization, multichip module packages, and higher levels of integration of integrated circuits (ICs) and MEMS1,2. The use of different materials, such as silicon (Si), for MEMS and ICs chips and polymer laminates for printed circuit boards (PCB) in a multichip module, results in major challenges caused by the mismatch of the coefcients of thermal expansion (CTE) of the involved materials, different I/O densities between the components, and resulting stresses, as depicted in Figure 1a3. To overcome these problems, silicon interposer substrates with through-silicon vias (TSVs) can be introduced as an interface between the PCB and the IC and MEMS dies, as illustrated in Figure 1b4. An interposer with TSVs provides electrical connections through the interposer substrate and enables vertical stacking of the chips, shorter interconnect lengths, reduced size of the module, as well as improved energy efciency of the module2,5. The interposer provides a possibility to integrate dies that are manufactured with different technologies, such as memory, power, logic, MEMS, and radio frequency (RF). The use of interposer substrates with TSVs is often referred to as2.5D integration2,4.

TSVs with electrically conductive Si via cores have been realized by several MEMS manufacturers and foundries6. For these Si TSVs, trenches are etched in a heavily doped Si substrate, insulated with SiO2, and then lled with non-conductive poly-Si. By using TSVs

with electrically conducting cores made of single-crystalline Si, a perfect match of the CTE values of the via cores and the Si substrate is obtained, thereby effectively eliminating potential reliability issues. However, Si-based TSVs have resistances in the range of 0.51 and thus can only be used for low-frequency applications because the high resistance in the TSVs causes signicant resistance-capacitance (RC) delay and signal decay.

Thus, to obtain high electrical conductivity (low resistance) and thereby low losses for RF applications or other applications, TSVs with conductive metal cores, for example, made of Cu or Au, are needed2. The drawbacks of incorporating such metals are the reliability issues due to the different CTE values of the involved materials, for example, Cu or Au and Si. The difference in the CTE values of Cu (16.7 p.p.m. C1) and Si (2.5 p.p.m. C1)5 will induce mechanical stress at the material interfaces7. If vias are completely lled with Cu, then cracks in the Si can emerge during heating8. To avoid the formation of cracks in the Si substrate, the inner walls of the TSV holes can be conformally coated by conventional Cu plating, thereby creating hollow metal TSVs (Figures 2af)9. Thus, the empty space at the center of the vias allows the Cu to expand when heated, thereby minimizing the risk of crack formation in the Si substrate. In addition, a conformal Cu lm limits the time consuming and expensive Cu plating, and reduces material consumption. Such a TSV fabrication process (Figures 2ai) was for example reported in Refs. 9 and 10, where the Cu-metalized TSVs are nally coated with a thin layer of benzocyclobutene (BCB) to achieve passivation and to realize a masking layer for the under bump metallization (UBM) layer. The UBM pads were placed next

1Department of Electronics and Communications Engineering, Tampere University of Technology, Korkeakoulunkatu 3, Tampere 33720, Finland; 2Silex Microsystems AB, Bruttovgen 1, Jrflla/Stockholm SE-175 26, Sweden; 3Micro and Nanosystem, KTH Royal Institute of Technology, Osquldas vg 10, 10044 Stockholm, Sweden; 4EMSL, Chalmers University of Technology, Kemivgen 9, 41296 Gteborg, Sweden and 5Mycronic AB, Nytorpsvgen 9, 18303 Tby, SwedenCorrespondence: Behnam Khorramdel (mailto:[email protected]

Web End =behnam.khorramdel@tut. ) or Frank Niklaus (mailto:[email protected]

Web End [email protected]) or Matti Mntysalo (mailto:[email protected]

Web End =matti.mantysalo@tut. )

Received: 31 August 2016; revised: 25 November 2016; accepted: 6 December 2016

Inkjet printing for increasing I/O density

B Khorramdel et al

2

a b

Components Coefficient of thermal expansion

Interface connection

DIE substrate

Nano/ASIC

ASIC

TSV interposer

BGA

ASIC

Rigid TSV

interposer

CTE 35 ppm C

Interposer Packaging substrate

Human interface

Micro-bump C4 bump Ball grid array

Connectors

CTE 58 ppm C

CTE 1217 ppm C CTE 1822 ppm C

Printed circuit board

PCB PCB

Figure 1 (a) Interface pyramid showing the components in a conventional package, along with the estimated CTEs of the involved substrates. (b) Introducing a rigid 3D TSV interposer (right) to bridge the mismatch of the CTE values of the PCB board and the Si application-specic integrated circuit (ASIC) dies3. CTE, coefcients of thermal expansion; PCB, printed circuit boards; 3D TSV, three dimensional through silicon via.

Si

SiO2 SixNy

TiW / Cu Cu Polymer Ag or Au SnAgCu

Resist

Fabrication of metalized TSVs

a

DRIE BS

d

Insulation and barrier

Seed layer and double-sided Cu plating

Seed layer removal

180 m 75 m

305 m

200 m

25 m

b

e

DRIE FS

DRIE FS tapered profile

c f

8 m

BCB coated vias with UBM pad at the via side

Dielectric filled vias with UBM pad on top of the via

g

BCB spray coating

UBM pad by electroless Ni/Au plating

Solder ball attachment

j

Dielectric filling using inkjet or super inkjet (SIJ)

UBM pad using inkjet or SIJ printing with Ag or Au ink

Solder ball attachment

h

k

i

l

Figure 2 (af) Process ow for the TSV fabrication9. (gi) Conventional fabrication approach with solder balls placed next to the via holes, resulting in low I/O densities, as in Ref. 10. (jl) Approach demonstrated in this work with TSVs lled with inkjet-printed dielectric polymer and solder balls placed in the same area as the TSVs to increase the I/O density. I/O, input/output; TSV, through silicon via.

Microsystems & Nanoengineering doi:http://dx.doi.org/10.1038/micronano.2017.2

Web End =10.1038/micronano.2017.2

Inkjet printing for increasing I/O density B Khorramdel et al

3

to the TSVs, followed by solder ball attachment, creating a solder ball pitch of 1 mm. By this method, it is possible to fabricate one I/ O per mm2 with proven reliability when exposed to temperature cycling (40 C to 125 C) of more than 1000 cycles8. However, the drawback of creating hollow TSVs is that they require a comparably large foot print, resulting in low I/O densities. There are two potential approaches to tackle the increased foot print: the diameter of the TSVs can be reduced and/or the UBM pads and the solder balls can be placed directly on top of the TSVs. By decreasing the TSV diameter, the aspect ratio of the vias will be increased, and practical limitations for conformal metal seed layer deposition inside the high-aspect ratio via holes by physical vapor deposition (PVD) or chemical vapor deposition (CVD) technologies are quickly reached. More complex and expensive deposition techniques, such as metalorganic-CVD or atomic layer deposition (ALD), are then required11. An alternative metallization method may be to use additive technologies, such as metal inkjet printing, to ll the high-aspect ratio TSVs. This technology requires fewer process steps as well as less waste material compared to conventional processes, such as electroplating. Inkjet printing of planar structures, such as conductive traces and large area devices (for example, antennas, electrodes, and photovoltaics), has successfully been demonstrated. However, in the case of partial metallization of the vias, metal inkjet printing technology is not sufciently mature (especially with Cu ink) for volume production and therefore not a viable solution for via metallization1218. An

alternate approach is to take advantage of the empty space in the Cu-plated hollow TSVs and place the UBM pads and the solder bumps on top of the TSVs, as shown in Figures 2jl, instead of placing the bumps next to the TSVs, as described in Ref. 10. For this purpose, the hollow vias must be lled with a suitable dielectric material with a low CTE to match the CTE of the Si. Previously, polymer lling of TSVs using spin coating of liquid polymers and vacuum-assisted lamination of dry lm resists have been investigated in Refs. 1921. The drawback with using spin-coating or vacuum lamination of polymers is the extensive optimization required to remove air bubbles that can become trapped in the polymer, potentially causing reliability problems. In addition, the topography of wafers with hollow TSVs strongly depends on the via density, thereby further complicating process optimization of the coating processes. Using inkjet printing, each via is lled separately, thereby eliminating the risk of non-uniform coatings over the wafer. Furthermore, targeted drop-by-drop lling with inkjet technology in combination with vacuum drying reduces the risk of void formation.

Therefore, in this work, we demonstrate the lling of hollow metal TSVs for interposer applications by two different printing methods, that is, inkjet and super inkjet (SIJ), using an inkjettable UV-curable hybrid polymer. We study cross sections of the resulting TSVs and electrically tested the lled TSVs before and after temperature cycling stress testing. Furthermore, we study the feasibility of printing metal pads, consisting of either Ag or Au, on top of the dielectric layer as the UBM layer. In addition, solder balls were dispensed on the UBM pads as a proof of concept. The proposed approach, compared to the technology used in Ref. 10, enables four times higher I/O densities without using the next-generation TSV node. This has signicant advantages because moving to the next-generation TSV node would involve smaller TSV dimensions with signicantly increased fabrication challenges.

MATERIALS AND METHODSFabrication of metalized TSVsThe fabrication of the metalized TSVs was executed on 200 mm diameter and 305-m-thick Si wafers (310 k cm). The use of rigid Si interposers that are 300430 m thick avoids the challenges involved in handling ultrathin wafers, including

the steps of bonding and debonding of the interposer substrate to temporary carrier wafers2. In addition, a thicker interposer substrate will better accommodate the stresses and reduce the stress gradient across the substrate, thereby reducing the wafer bow3. The via holes are etched by deep reaction ion etching on both the frontside (FS) and the backside (BS) of the wafer, as shown in Figures 2a and b. To be able to close the via holes and create hermetically sealed TSVs by plating, the FS vias are etched with tapered structure, as shown in Figure 2c. The FS vias are 25 10 m deep, the surface openings are 25 5 m in diameter, and the bottom openings of the vias on the FS are 8 3 m in diameter. SiO2 is used as an insulation layer, SixNy as a wafer bow stress compensation layer, and a TiW/Cu layer is used as a barrier and seed layer, as shown in Figure 2d5. A Cu-plating process was used to ll the FS vias and, in the same step, conformally plate a layer of Cu on the inner walls of the BS via holes. Figure 2e shows the Cu redistribution layer (RDL) on both the FS and BS of the wafer, which was plated in the same Cu-plating step. This combined Cu deposition process for both the via ll and the RDL eliminates metal interfaces between the vias and the RDL, which otherwise can cause poor electrical contacts and conductive losses, especially in RF applications. In the nal step, the resist and the Cu/TiW seed layer are removed, as illustrated in Figure 2f9.

Material selection for TSV llingTo select a suitable dielectric material for lling the TSVs, four critical key material parameters must be considered for the target application: (1) Low-k to obtain good RF properties, even at frequencies in the GHz range, (2) low Youngs modulus to minimize the introduced stress levels, (3) CTE of the dielectric that is as close as possible to the CTE of Si and Cu to minimize the induced stress levels in the TSV interposer during the soldering process and during temperature cycling, and (4) low material shrinkage during the curing process. Considering all of these factors, we selected InkOrmo (micro resist technology GmbH, Berlin, Germany), which is an inkjettable UV-curable hybrid polymer with a low Youngs modulus of ~ 1 GPa, low-volume shrinkage and high thermal stability of up to 300 C (short term) and 270 C (long term), which is higher than the typical bonding temperature (217 C) for the Sn96.5%Ag3%Cu0.5%

alloy paste. In addition, the heat resistance of the InkOrmo polymer is higher than typical peak temperatures (250 C) of reow processes used with the SnAgCu alloys. The low Youngs modulus of the polymer (~100 times less than Cu-plated coverage) results in substantially reduced thermal stresses in the interposer compared to the thermal stresses induced by typical vias that are completely lled with a metal, despite that the CTE of the InkOrmo polymer is not perfectly matched to the substrate material. Table 1 presents the detailed specications of the InkOrmo material.

Table 1 Material specications for the polymer dielectric InkOrmo

Material specicationsLiquid polymer pre-cursor before curing

Viscosity (mPa s) 12 Hybrid polymer after photocuring ( = 365 nm) and hardbake (140160 C)

Volume shrinkage during UV curing (%) 57 Volume shrinkage during hardbake (%) 12 CTE (20100 C) (p.p.m. K1) 60

Youngs modulus (GPa) ~ 1 Hardness (by indentation) (MPa) ~ 68

doi:http://dx.doi.org/10.1038/micronano.2017.2

Web End =10.1038/micronano.2017.2

Microsystems & Nanoengineering

Inkjet printing for increasing I/O density B Khorramdel et al

4

Table 2 Temperature cycling standards and specications

Standard conditions Specications

JEDEC spec JESD22-A104 Condition G 40 C to +125 C Temperature cycle 2 times per hour Duration 1000 cycles Soak time 1314 min

TSV lling and UBM deposition with inkjetA Dimatix DMP-2800 inkjet printer (Fujilm, Lebanon, New

Hampshire, USA) equipped with a nominal 10 pL cartridge (DMC-11610) was used for printing the precursor of the polymer dielectric into the via holes. The actual inner diameter of the BS via holes is 180 m after the 10 m Cu liner deposition, and the depth of 270 m. For the printing and lling trials, a square pattern including 11 11 droplets per pixel (121 in total) with a resolution of 5080 dpi (drop space of 5 m) was prepared.

The interposer structure was ushed with isopropanol and heated to 200 C for 5 min and then cooled to room temperature before drop deposition as recommended by the ink supplier. First, a total of 20 layers were printed in two rounds with drying of the ink (evaporation of solvent) in a vacuum oven after each round. The printing parameters are listed in Supplementary Table S1. During the lling process, the printing parameters were stable, and the jetting direction of the droplets was straight over time. Initial experiments showed that the printed material volume was not sufcient for complete lling of the hollow TSVs. Thus, instead of two printing rounds, 26 layers were printed in three rounds (10 +10+6) with two phases of drying in a vacuum oven (30 min at 65 C) after the rst and second rounds of printing. The 26 layers are equal to ~ 3000 droplets, which can be compared to the theoretically calculated 700 droplets comprising the polymer precursor and solvent required to ll the vias. This difference shows the effect of evaporation of the solvent from the polymer precursor and shrinkage of the polymer because of baking and curing of the material in the subsequent lling rounds. The lowest available pressure (102 mbar) was used in the vacuum oven to avoid the formation of pockets with trapped air inside the via holes during the drying steps (solvent evaporation). After the nal drying step, the dielectric polymer in the via holes was UV-cured for ~ 20 s before heating the sample to 150 C for 3 h on a hotplate to hardbake the polymer. During printing of the polymer into a single via hole, there was no delay between printing the rst layers; however, during printing of the last layers, a delay was required for solvent evaporation. Because printing one layer by lling the whole TSV interposer with tens of vias takes 10 min and 40 s, no additional delay was required for solvent evaporation.

After lling the vias with the dielectric, a conventional inkjet was used to print Ag UBM pads on top of the vias using an established and reliable Ag ink, NPS-J silver ink from Harima Chemicals, Tokyo, Japan (particle size: 12 nm, metal content: 65 wt% and solvent: Tetradecane), as a proof of concept. The pads were circular with a diameter of 300 m, overlapping 75 m with the Cu collar of the TSV for connection. Ni-based ink may be a better choice because there is a risk for dissolving Ag pads into the solder balls during solder ball formation and ball bonding. However, at the time, no suitable inkjettable Ni ink was available for this experiment. Copper ink could be another alternative option; however, in our experience, Cu annealing processes are relatively complex, typically requiring photonic or thermal annealing in inert atmosphere with forming gas22,23.

TSV lling and UBM deposition with SIJAs a potential alternative, SIJ24 (SIJ-S050) manufactured by

SIJTechnology, Inc. (Tsukuba, Japan) was also investigated for dispensing the dielectric material inside the via holes using a needle-shaped nozzle. SIJ is a commercial electrohydrodynamic inkjet printing technology that allows for sub-femtoliter droplet deposition with high precision. In our SIJ set-up, the nozzle is placed over a via hole to dispense the dielectric material and completely ll the hole with the dielectric before the nozzle is moved to the next hole. Because of the shrinkage of the dielectric polymer during the prebake in the vacuum oven, lling is performed in several rounds until the hollow TSV is completely lled. After nishing the lling process, the nal prebake (6090 C

for 1530 min), UV curing (5001500 mJ cm2), and hard baking (150 C for 3 h) was executed. Because Ag UBM pads printed by inkjet printing showed non-uniform coverage on top of the TSVs, SIJ was used to dispense Au ink during a spiral movement of the nozzle on top of the TSVs, starting at the edge of the copper collar and moving to the center. The detailed printing parameters are listed in Supplementary Table S2. The UBM pads were printed with Au ink (UTDots Inc., Champaign, Illinois, USA) and subsequently sintered in an oven at 250 C for 1 h.

Reliability testing by temperature cyclingTemperature cycling reliability testing was performed after the dielectric lling of the metal TSVs to benchmark against the initial reliability tests performed on the standard hollow metal TSVs10.

The TSVs used for our experiments are identical (taken from the same wafer batch) to the TSVs used in Ref. 9. The resistances of these TSVs were measured to be on the order of 10.3 m per via before lling the hollow vias with the dielectric polymer and placing the metal on top of the lling. The insulator does not affect the resistance, and the metal coverage thickness is two orders of magnitude smaller than the via structure. Thus, by its very nature, these additions have a minimal effect on the via resistance.

The resistance of a via daisy chain was measured before and after the temperature cycling stress test (see Table 2) after 250, 500, 750, and 1000 cycles using a Keithley 2400 sourcemeter attached to a probe station (two-probe DC measurement). The overall measured resistance was ~ 2 , which includes six vias in the chain, Cu RDLs connecting the vias and measurement pads, contact resistances, and cable resistances of the measurement setup. The cable had the highest contribution to the overall resistance. However, the main purpose with the two-probe measurement was to examine the via chains for possible open circuits caused by damage of the plated copper after the temperature cycling reliability test. Figure 3 illustrates the daisy chain with pads on the left and right side designated for landing the probes to perform the electrical measurements.

After the electrical measurements, one via from each sample (three vias in total) was selected for preparing via cross sections using broad ion beam (BIB) technology to introduce the least possible damage to the samples. The cross sections were milled using 6 kV argon ions and an ~ 100-m-thick layer was removed (half of the via). Next, scanning electron microscope (SEM) was used to characterize the cross sections to determine how the temperature cycling affects the dielectric lling inside the TSVs.

For the SIJ-printed Au UBM pads on top of the vias, the same reliability test was performed. The sample was imaged by SEM after 250 and 500 temperature cycles to evaluate the effect of temperature cycling on the pads.

FS Cu RDL BS Cu RDL

Figure 3 Via daisy chain structure. Two probes land on the probing pads on the left and the right side of the via chain.

Microsystems & Nanoengineering doi:http://dx.doi.org/10.1038/micronano.2017.2

Web End =10.1038/micronano.2017.2

Inkjet printing for increasing I/O density B Khorramdel et al

5

a b

Si

Si

Polymer

Polymer

Cu

Cu

Ag

Figure 4 (a) Cross-sectional image of the TSV after successful dielectric lling by inkjet printing, where the white arrow at the bottom shows the surface of the dielectric layer. (b) Cross-sectional image of the TSV with the Ag UBM layer on top of the dielectric material. Scale bars are 50 m. TSV, through silicon via; UBM, under bump metallization.

Si

Solder ball attachmentThe solder ball attachment on the Ag UBM pads was performed using a novel jetting technology (MY600 Jetprinter, Mycronic AB, Tby, Sweden) developed specically for high-viscosity functional uids. The technology utilizes a piezo-induced volumetric displacement that forces uid out through a single 100 m-wide nozzle at jetting frequencies of up to 300 Hz. The uid in this case was a SnAgCu 305 solder paste (96.5% Sn, 3% Ag, and 0.5% Cu; from Senju Metal, Senju Hashido-cho 23, Adachi-ku, Tokyo 120-8555, Japan). The paste consists of spherical metallic alloy granules together with an organic resin-based uid binder in an 8218 wt% mixture. Droplets with a goal diameter of 227 m and approximate volume of 2 nL were deposited on the previously deposited Au UBM layers. Positioning accuracy of the deposits is 12 m (standard deviation). To form a ball connection, the solder paste was reowed in a commercial vapor phase oven (ASSCON VP800, ASSCON Systemtechnik Elektronik GmbH, Messerschmittring 35, D-86343 Knigsbrunn, Germany) utilizing a maximum vapor temperature of 230 C with a temperature gradient of0.7 C s1 between 50 and 100 C, 1.8 C s1 between 100200 C and 0.8 C s1 from 200 C to the peak temperature.

RESULTS AND DISCUSSIONDielectric lling and Ag UBM pads by inkjet printingLight microscope images of the hollow copper TSVs that were lled by inkjet printing, after the nal baking, showed a sufcient lling level without any voids but with a small concave prole at the surface, as depicted in Figure 4a. The UBM printing results using inkjet technology, as presented in Figure 4b, showed that the silver ink tends to move toward the center of the TSV, resulting in areas on the via edges that are not completely covered with ink and thus may result in reduced electrical connectivity to the plated copper. The minor difference in via depth that is visible in Figures 4a and b is attributed to etch rate variations within a wafer batch during TSV fabrication. The prole of the TSVs appears different in Figures 4a and b; this difference is an artifact resulting from slight tilting and off-center placement

of the TSV cross sections. The peeling of the metallization at the bottom of the via in Figure 4b is caused by the preparation of the cross section, that is, by the grinding step. Furthermore, the reason that the connection to the FS RDL is not visible in Figure 4 is that the cross sections are not exactly at the center of the vias because of the limitations in the manual preparation of the cross sections.

Dielectric lling and Au UBM pad by SIJ printingAs mentioned in the previous section, SIJ printing was evaluated as an alternative for lling of the via holes. In Figure 5, an SEM

Cu

Polymer

Figure 5 SEM image of a via hole lled using the dispensing mode of super inkjet (SIJ). Scale bar is 20 m. SEM, scanning electron microscope.

doi:http://dx.doi.org/10.1038/micronano.2017.2

Web End =10.1038/micronano.2017.2

Microsystems & Nanoengineering

Inkjet printing for increasing I/O density B Khorramdel et al

6

a b

Au UBM

Figure 6 SEM image of (a) the top view of the interposer with empty vias and the adjacent pads. (b) Via lled with the dielectric and Au UBM dispensed on top by SIJ. Scales are 100 and 50 m, respectively. SEM, scanning electron microscope; SIJ, super inkjet; UBM, under bump metallization.

Plated copper

a b c

Baked dielectric material

Adhesion issue

After 250 cycles After 750 cycles After 1000 cycles

Figure 7 BIB prepared cross-section SEM images of the lled vias after temperature cycling tests with (a) 250 cycles, (b) 750 cycles, and (c) 1000 cycles. Scale bars are 20 m. In (b), the connection between the via metallization and the FS RDL on the top wafer surface cannot be seen because the cross section is not exactly passing through the center of the via. BIB, broad ion beam; FS RDL, frontside redistribution layer;

SEM, scanning electron microscope.

micrograph of the cross section of a single via lled by the dispensing mode of SIJ printing is shown. It was concluded that this method can be even faster than using inkjet printing to ll the via holes in applications in which a small number of vias must be lled. In these cases, the lling level of the via hole and the planarity of the dielectric polymer at the via hole surface was improved compared to via holes that were lled by inkjet printing. Furthermore, no voids or other anomalies were observed. However, at the bottom and on one side wall of the via holes, detachment of the dielectric polymer was observed. This could be caused by adhesion problems caused by impurities and moisture as a result of insufcient substrate cleaning (ushing with isopropanol and drying on a hotplate). Poor adhesion could be improved and optimized by adhesion promoters.

By printing circular Au UBM pads using SIJ, complete coverage of the top surface of the lled vias was obtained with a proper connection to the Cu collar around the edge of the vias. In Figure 6a, a dashed circle representing the SEM image of a via with no dielectric lling is shown, and Figure 6b depicts a via lled with the dielectric and Au UBM dispensed on top by SIJ printing, as indicated by the dashed circle.

Temperature cycle stress testElectrical testing before and after the temperature cycling (250, 500, 750, and 1000 cycles) was performed using a two-probe DC measurement set-up. No open circuits or signicant resistance changes were observed.

Figure 7 illustrates cross sections of the vias lled by SIJ printing with the dielectric polymer after temperature cycling tests with (a) 250 cycles, (b) 750 cycles, and (c) 1000 cycles. Shrinkage and detachment of the dielectric polymer lling was observed; the shrinkage and detachment increased with an increasing number of temperature cycles. In addition, small micro-cracks in the polymer can be seen on the bottom-right corner of the vias in Figures 7a and b. These defects could be due to the adhesion problems, polymer shrinkage, or the mismatch between the CTEs of the polymer and the substrate material. No anomalies in the SiCu interfaces were observed.

Figure 8 shows SEM images of two different samples with Au UBM pads before and after temperature cycling. In Sample 1, after 250 temperature cycles, micro-cracks were observed in a circular pattern at the interface between the Cu and the dielectric polymer (Figure 8b). This failure is likely the same type of micro-crack as

Microsystems & Nanoengineering doi:http://dx.doi.org/10.1038/micronano.2017.2

Web End =10.1038/micronano.2017.2

Inkjet printing for increasing I/O density B Khorramdel et al

7

a b c

After 250 cycles

After 500 cycles

d e f

After 250 cycles After 500 cycles

Figure 8 Top view SEM images of the SIJ printed Au UBM pads. Sample 1: (a) before temperature cycling and after temperature cycling with (b) 250 cycles and (c) 500 cycles. Sample 2: (d) before temperature cycling and after temperature cycling with (e) 250 cycles and (f) 500 cycles. Scale bars are 50 m. SEM, scanning electron microscope; SIJ, super inkjet; UBM, under bump metallization.

that observed in Figures 7a and b. Adding another 250 temperature cycles did not cause any additional defects (Figure 8c). The same observation applies to Sample 2 (Figure 8d), where one micro-crack with the size of ~ 50 m was visible after the rst 250 cycles (Figure 8e), and no additional visible defects were observed after adding another 250 temperature cycles (Figure 8f).

Solder ball attachmentThe solder ball attachment experiments showed successful jetting of the SnAgCu paste on the Au UBM pads, as depicted in Figure 9a. The volume distribution over a series of ball grid arrays using the MY600 Jetprinter depends on the target ball diameter, which is ~ 510% (1 sigma). The stated positioning accuracy is a reliable measure and is sufcient for PCB features with pad sizes as small as 200 m and pitches as small as under 350 m. Figure 9b illustrates the dramatic increase in I/O density enabled by our approach, without the need for using next-generation TSV with reduced diameters. The conventional approach10 (Figures 2gi and 9b (left)) allows for a pitch of the solder balls of 1000 m, resulting in 1 I/Os per mm2. The novel approach presented in this work (Figures 2jl and 9b (right)) enables placing four solder balls per mm2 (I/Os per mm2), resulting in four times higher I/O density while retaining identical TSV dimensions.

CONCLUSIONIn this work, inkjet printing and SIJ technology were demonstrated to be available for use in performing the process of dielectric lling of hollow TSVs with an UV-curable hybrid polymer that has a

low Youngs modulus. These inkjet methods have been optimized to completely ll the via holes without any voids. The lling process with SIJ produced almost planar surfaces, whereas inkjet-lled vias suffered from a slight concave prole. However, after baking, SIJ-lled vias suffered from partial detachment of the polymer from the via walls, which could be attributed to weak adhesion. It was concluded that the SIJ technology is a better alternative compared to inkjet printing regarding dispensing of metal inks for the UBM pads on top of the vias lled with a dielectric polymer.

Electrical testing before and after temperature cycling experiments showed that temperature cycles of up to 1000 cycles did not cause any open circuits or signicantly affect the electrical resistance of the via chains with printed dielectric polymer lling, conrming the ndings in Ref. 10. Thus, the difference of the CTEs of Si, Cu, and the dielectric polymer do not appear to affect the electrical functionality of the TSVs. By studying cross sections of the lled vias after temperature cycling, shrinkage and adhesion problems between the dielectric polymer and the bottom and the side walls of the via holes were observed, causing partial detachment of the dielectric polymer in these areas. This effect increased with increasing numbers of temperature cycles (up to 1000 cycles). Dielectric polymers with lower volume shrinkage and a CTE that is better matched to the CTE of the substrate material may be introduced to reduce the observed detachment of the polymer from the sidewalls of the via holes. It was also demonstrated that solder balls can be printed successfully on the interposers, thereby demonstrating the potential for a four-times increased I/O density compared to the same TSV platform used in Ref. 10 but without lling of the hollow vias with the dielectric polymer.

doi:http://dx.doi.org/10.1038/micronano.2017.2

Web End =10.1038/micronano.2017.2

Microsystems & Nanoengineering

Inkjet printing for increasing I/O density B Khorramdel et al

8

a

b

1000

350

635

365

500 350

200

150

200

1000

Figure 9 (a) SEM images of solder paste deposited on the vias lled with dielectric polymer and covered by printed UBM. Scale bars are 50 m. (b) Schematic of the UBM and solder ball layout on the interposer for demonstrating the increase in I/O density. I/O, input/output;

SEM, scanning electron microscope; UBM, under bump metallization.

ACKNOWLEDGEMENTS

This work is supported by ENIAC-JU Project Prominent Grant No 324189 and Tekes Grant No. 40336/12 and Vinnova Grants Nos. 201204301, 201204287, and 201204314. MM is supported by the Academy of Finland Grant Nos. 288945 and 294119. The work of Silex and KTH was funded in part through an Industrial Ph.D. grant from the Swedish Foundation for Strategic Research (SSF), Grant No. ID140033. The authors acknowledge Okmetic for providing Si wafers needed for the fabrication of metallized TSVs. We also thank Jere Manni from Top Analytica for making the cross sections of the lled vias using BIB technique and Janne Kiilunen from Tampere University of Technology, Reliability Group for performing the temperature cycling tests. Lucas Held at Silex is greatly acknowledged for design of the TSV interposer, the wafer level interconnect design and CAD support. We also thank the fabrication personnel in the Silex 8-inch fab for assisting with processing the TSV interposer wafers.

COMPETING INTERESTS

The authors declare no conict of interest.

REFERENCES

1 Ogutu P, Fey E, Dimitrov N. Superconformal lling of high aspect ratio through glass vias (TGV) for interposer applications using TNBT and NTBC additives. Journal of the Electrochemical Society 2015; 162: D457D464.

2 Wang Z. 3-D integration and through-silicon vias in MEMS and microsensors. Journal of Microelectromechanical Systems 2015; 24: 12111244.

3 Ebefors T, Perttu D. CTE matched interposer and method of making. US Patent No. 9224681. 2015. Available at https://www.google.com/patents/US9224681

Web End =https://www.google.com/patents/US9224681 .4 Fischer AC, Forsberg F, Lapisa M et al. Integrating MEMS and ICs. Microsystems & Nanoengineering 2015; 1: 15005.

5 Malta D. TSV formation overview. In: Garrou P, Koyanagi M, Ramm P (eds). Handbook of 3D Integration. Wiley-VCH Verlag GmbH & Co. KGaA: Weinheim, Germany, 2014: 6578.

6 Klvesten E, Ebefors T, Svedin N et al. Electrical connections in substrates. US Patent No. 7560802. 2009. Available at https://www.google.com/patents/US7560802

Web End =https://www.google.com/patents/ https://www.google.com/patents/US7560802

Web End =US7560802 .

7 Asiatici M, Laakso MJ, Fischer AC et al. Niklaus F. Through silicon vias with invar metal conductor for high-temperature applications. Journal of Microelectromechanical Systems 2017; 26: 158168.

8 Frank T, Chappaz C, Leduc P et al. Reliability approach of high density through silicon via (TSV). 12th Electronics Packaging Technology Conference; 810 Dec 2010; Singapore; 2010: 321324.

9 Ebefors T, Fredlund J, Jung E et al. Recent results using met-via TSV interposer technology as TMV element in wafer-level through mold via packaging of CMOS biosensors. International Wafer-Level Packaging Conference (IWLPC); 67 Nov 2013; San Jose, CA, USA; 2013: 18.

10 Yazdani F. Design and direct assembly of 2.5D/3D rigid silicon interposer on PCB. International Symposium on Microelectronics 1316 Oct 2014; San Diego, CA, USA; 2014: 783786.

11 Ebefors T, Fredlund J, Perttu D et al. The development and evaluation of RF TSV for 3D IPD applications. 2013 IEEE International 3D Systems Integration Conference (3DIC); 24 Oct 2013; San Francisco, CA, USA; 2013: 18. doi: 10.1109/ 3DIC.2013.6702382.

12 Rathjen A, Bergmann Y, Krger K. Feasibility study: Inkjet lling of through silicon vias (TSV). NIP & Digital Fabrication Conference; 913 Sep 2012; Quebec City, Canada; 2012: 456460.

13 Khorramdel B, Mntysalo M. Fabrication and electrical characterization of partially metallized vias fabricated by inkjet. Journal of Micromechanics and Micro-engineering 2016; 26: 45017.

14 Khorramdel B, Laurila MM, Mantysalo M. Metallization of high density TSVs using super inkjet technology. IEEE 65th Electronic Components and Technology Conference (ECTC); 2629 May 2015; San Diego, CA, USA; 2015: 4145.

15 Khorramdel B, Mantysalo M. Inkjet lling of TSVs with silver nanoparticle ink. Proceedings of the 5th Electronics System-Integration Technology Conference (ESTC); 1618 Sep 2014; Helsinki, Finland; 2014: 15.

16 Eiroma K, Viljanen H. Application of inkjet printing for 3D integration. NIP & Digital Fabrication Conference; 28 Sep1 Oct 2015; Protland, OR, USA; 2015: 195200.

17 Quack N, Sadie J, Subramanian V et al. Through silicon vias and thermo-compression bonding using inkjet-printed gold nanoparticles for heterogeneous MEMS integration. Transducers & Eurosensors XXVII: The 17th International Conference on Solid-State Sensors, Actuators and Microsystems (TRANSDUCERS & EUROSENSORS XXVII); 1620 Jun 2013; Barcelona, Spain; 2013: 834837.

18 Sadie J, Quack N, Wu MC et al. Droplet-on-demand Inkjet-lled through-silicon Vias (TSVs) as a pathway to cost-efcient chip stacking. 46th International Symposium on Microelctronics; 30 Sep3 Oct 2013; Orlando, FL, USA; 2013: 866871.

19 Bouchoucha M, Chausse P, Moreau S et al. Reliability study of 3D-WLP through silicon via with innovative polymer lling integration. IEEE 61st Electronic

Microsystems & Nanoengineering doi:http://dx.doi.org/10.1038/micronano.2017.2

Web End =10.1038/micronano.2017.2

Inkjet printing for increasing I/O density B Khorramdel et al

9

24 Murata K, Masuda K. Super inkjet printer technology and its properties.

Convertech e-Print 2011; 1: 108111.

This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the articles Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/

Web End =http://creativecommons.org/licenses/ http://creativecommons.org/licenses/by/4.0/

Web End =by/4.0/

The Author(s) 2017

Supplementary Information for this article can be found on the Microsystems & Nanoengineering website (http://www.nature.com/ micronano)

Components and Technology Conference (ECTC); 31 May3 Jun 2011; Lake Buena Vista, FL, USA; 2011: 567572.20 Bouchoucha M, Chapelon L-L, Chausse P et al. Through silicon via polymer lling for 3D-WLP applications. 3rd Electronics System-Integration Technology Conference (ESTC); 1316 Sep 2010; Berlin, Germany; 2010: 14; doi:10.1109/ ESTC.2010.5642998.21 Chausse P, Bouchoucha M, Henry D et al. Polymer lling of medium density through silicon via for 3D-packaging. 11th Electronics Packaging Technology Conference; 911 Sep 2009; Singapore; 2009: 790794.22 Soltani A, Khorramdel Vahed B, Mardoukhi A et al. Laser sintering of copper nanoparticles on top of silicon substrates. Nanotechnology 2015; 27: 35203.23 Niittynen J, Sowade E, Kang H et al. Comparison of laser and intense pulsed light sintering (IPL) for inkjet-printed copper nanoparticle layers. Scientic Reports 2015; 5: 8832.

doi:http://dx.doi.org/10.1038/micronano.2017.2

Web End =10.1038/micronano.2017.2

Microsystems & Nanoengineering