Novel Copper Metallization Schemes on Ultra-Thin, Bare Glass Interposers with Through-Vias

Timothy Huang, Bruce Chou, Venky Sundaram, Himani Sharma, and Rao Tummala

3D Systems Packaging Research Center

Georgia Institute of Technology

813 Ferst Drive, N.W.

Atlanta, GA USA 30332

Email: tim.huang@gatech.edu

Abstract

Metallizing ultra-thin glass interposer with through-vias

with high adhesion and at low cost is one of the primary

challenges in producing next-generation glass-based system

packages. This paper describes and investigates a new

approach towards creating a glass interposer structure with

through-vias that is ready for solution-based metallization such

as electroless deposition. Starting with glass containing

through-vias, a thin polymer film (primer) is laminated,

covering the entire surface. The film is then opened over the

vias, resulting in a structure that is ready for electroless

deposition and is known to be thermo-mechanically reliable.

The versatility and feasibility of this approach are demonstrated

through the use of various primer film materials and primer

opening processes. Daisy-chain reliability structures were

fabricated on glass interposers metallized by this approach and

electrical measurements showed expected behavior.

Introduction

Many recent advances in glass interposer manufacturing

technologies make glass the prime candidate material for low-

loss, high-performance interposer substrates. As the core

substrate for next-generation 2.5D and 3D system packages,

fine-pitch copper through-vias are needed to support fine-pitch

fan-out for fine-pitch bumps. Traditionally, metallizing an

ultra-thin glass interposer starts by laminating a polymer film

on both sides of a blank glass substrate followed by through-

via formation and metallization by electroless plating

(Figure 1(a)) [1]. The polymer film provides mechanical

support and also acts as an adhesion layer between glass and

the metal. However, such films are incompatible with some

current glass via-forming processes. A metallization process

that is universally compatible with any glass via-formation

method therefore must begin with bare glass containing

through-vias as the starting substrate. Sputtering is a process

known to metallize bare glass surfaces with strong adhesion [2,

3], but the process alone is unsuitable for adequately

metallizing high-aspect ratio through-vias [4]. In contrast,

metallization by electroless deposition has much higher

capabilities of reaching deep aspect ratios in glass while also

keeping the processing cost low. This paper explores and

demonstrates a novel approach to metallizing copper on ultra-

thin, bare glass interposers with through-vias by depositing and

defining an ultra-thin, patternable polymer film onto bare glass

with through-vias, which maintains the advantages of

improved handling and good metal adhesion from the previous

process while also ensuring universal compatibility with glass

via formation methods.

Figure 1(b) illustrates the process flow of this approach.

The process begins by laminating a thin polymer film (primer)

over glass with through-vias, followed by patterning the primer

to open the through-vias. Next, the primer must be defined and

opened to expose the underlying glass vias. The feasibility and

versatility of this process will be demonstrated by utilizing

various primer materials and opening technologies. Finally, the

patterned primer and exposed through-vias can be metallized

with electroless copper, yielding a structure that is known to be

reliable [1].

Figure 1. Process flow diagrams comparing the (a) previously

established process with the (b) novel process.

Metallized through-vias formed by this method are

characterized to assess the process feasibility. The feasibility of

metallizing via arrays with 60µm diameter at 120µm pitch in

100-130µm thick glass as well as its conformality will be

determined. Finally, daisy-chain structures will be fabricated

and subjected to thermal cycling to assess the thermo-

mechanical performance. The fundamental challenges of each

process are assessed and future outlook is discussed.

Experimental Methods

For glass substrates with vias, 130µm thick SGW3

(Corning Inc.) and 100µm thick EN-A1 (Asahi Glass Co.) were

used. The substrates dimensions were 76.2 x 76.2 mm2. Sixteen

via arrays were defined on the substrate. Within each array

there were 16 x 16 vias, where all via diameters at the entrance

were specified to be 60µm at 120µm pitch. Two primer film

materials were used: Ajinomoto Buildup Film (hereby referred

to as “ABF”) GX-92P (Ajinomoto Co., Inc.) with 5µm

thickness and an experimental material by Zeon Corp with 3µm

thickness (hereby referred to as “ZF”).

978-1-4799-8609-5/15/$31.00 ©2015 IEEE 1208 2015 Electronic Components & Technology Conference

Prior to primer lamination, glass was prepared as follows.

Substrates were first cleaned with acetone and isopropyl

alcohol, followed by an O2 plasma clean. A 0.9% solution of

(3-Aminopropyl)triethoxysilane (99%, Sigma-Aldrich) in

ethanol was then coated on both sides of the glass and cured for

20 min at 115°C.

ABF and ZF films were then vacuum laminated onto glass

and cured as per manufacturer instructions. Vacuum lamination

conditions were optimized to minimize air bubbles and voids.

Plasma etching of ABF and ZF in regions over glass vias

was done by first laminating dry film photoresist whose

thickness was greater than the underlying primer film. After

exposure and development, the defined photoresist served as an

etch mask, having open regions above the glass vias. For

plasma etching, a mix of O2 (100 sccm) and CF4 (25 sccm) gas

was used at 100°C with an RF power of 400 W for 90 min per

side.

Laser drilling was performed by Micron Laser Technology.

For both ABF and ZF, two different lasers were used to open

the primer films. One was a CO2 laser (λ = 10.6µm) with a

maximum output of 80 W, which used 10 pulses at about 20%

power. The second was a UV laser (λ = 266 nm), which

required 1 pulse at 90% power.

After primer films were opened, about 400 nm of Cu seed

layer was conformally plated using a commercial electroless

bath (Atotech Inc.). The final daisy-chain structures were

fabricated by semi-additive processing with a final Cu

thickness of about 10µm. Finally, samples were annealed at

160°C in air for 1 h.

The design of experiment of the daisy-chain structures

consisted of Kelvin resistance structures with an incremental

number of vias, from 2 to 66. Shown in Figure 2 is a view of

the electrical design with eight daisy-chains. A summary of the

daisy-chain structures is listed in Table I. Daisy-chain

resistances were measured before and throughout thermal cycle

testing using the standard four-point method of driving a 50 mA

current source from one end of the chain to the other and

sensing the voltage drop using a multi-meter (Keithley

Instruments). Thermal cycle testing was performed according

to JEDEC Standard JESD22-A104D Condition B, having a

maximum and minimum temperature of 125°C and -55°C,

respectively. A soak time of 15 minutes was used at the

maximum and minimum temperatures.

Table I. List of Kelvin daisy-chains.

ID # of vias ID # of vias

D1 2 D5 36

D2 8 D6 46

D3 16 D7 56

D4 26 D8 66

Ex. D1, 2-via daisy-chain

Figure 2. Design of daisy-chain Kelvin resistance

measurement structures.

Results and Discussion

Glass substrates were successfully laminated with ABF and

ZF. It should be noted that in the cases of ABF and ZF,

lamination of such thin films initially resulted in incomplete

adhesion to the substrate, resulting in voids and air bubbles.

Only after optimizing vacuum lamination conditions were ABF

and ZF able to be laminated with high quality consistently

(Figure 3). Improvements in lamination conditions include

increasing the vacuum and pressure dwell durations.

Figure 3. Glass laminated with (a) ABF and (b) ZF. The glass

vias can be seen beneath the primer films.

ABF and ZF films were successfully etched away by

plasma to open the glass vias (Figure 4). In the process of

plasma opening, an unexpected challenge was met in

fabrication. Both sides of each sample were laminated with dry

film photoresist prior to exposure, and due to the transparency

of the primer and glass, exposure on one side (mask side) (Fig.

5(a)) resulted in an unintentional exposure of the photoresist on

the other side (back side) (Figure 5(b)). The back side thus

receives a sub-optimal pattern transfer from the mask. This can

be seen in Figure 5 when comparing the smooth, round edges

of the circle on the mask side with the jagged edges on the back

side.

16 x 16 array of vias

(a) (b)Covered via

Covered via

1209

Figure 4. (a) ABF and (b) ZF after plasma etching. The open

vias and glass surface around each via are visible.

One solution to the problem of unwanted simultaneous

exposure is to perform lithography steps sequentially, one side

at a time. Having one side laminated with photoresist induced

a small but tolerable amount of warpage in the substrate. While

this warpage was feasible for these processing dimensions,

results may vary for designs with lower dimensional tolerances.

Figure 5. Top: schematic illustrating simultaneous, double-

sided exposure due to material transparency. Bottom:

micrographs after development illustrating the difference in

pattern quality between the (a) mask side and (b) back side.

ABF and ZF after laser ablation can be seen in Figure 6. It

is apparent that relative to the UV laser, the CO2 laser was able

to ablate both primer films within a smaller diameter and with

a more controlled, circular shape. The inner diameter of the

CO2 ablated circle was measured to be about 80µm while that

of the UV ablated region was oblong with a diameter of 80µm

in the short axis and 100µm in the long axis. While both CO2

and UV-ablated samples have exposed regions of glass

between the via and the non-ablated primer, the larger exposed

area of blank glass around vias from UV-ablated samples is

expected to have weaker adhesion to subsequent metal layers,

and its implications on thermo-mechanical reliability will be

discussed later in this paper. From a manufacturing

processability standpoint, the control and quality demonstrated

by the CO2 laser make it the preferred choice.

Comparing the primer opening methods used, it is clear that

the laser processing route has several advantages over plasma

etching in terms of processability and in fabrication quality.

Laser processing can be performed on one side a time without

warpage concerns and with good patterning. For this reason,

only laser-processed samples continued to metallization.

Figure 6. Optical micrographs of (a) ABF and (b) ZF after

CO2 laser ablation and (c) ABF and (d) ZF after UV laser

ablation.

Cu coverage after electroless deposition was continuous

and without any observed blistering or adhesion issues on all

samples (Figure 7). Although the UV laser etched excess

primer material and left large regions of bare glass around the

vias, it did not appear to have any negative effects on the

adhesion to Cu in those regions (Figure 7(c, d)).

Figure 7. Optical micrographs after electroless plating on (a)

ABF and (b) ZF after CO2 laser ablation and (c) ABF and (d)

ZF after UV laser ablation.

Daisy-chain structures for thermal cycle testing were

successfully fabricated on all samples through semi-additive

processing (Figure 8). Although electrolytic metallization was

intended to conformally plate the via sidewalls, the samples

with ZF primer appear to have filled Cu vias (Figure 8(b, d)).

They are not Cu-filled vias, but only filled at the ends due to

insufficient turbulent flow in the electrolytic plating tank. The

UV-etched primer vias maintained sufficient Cu adhesion at the

daisy-chain pads, which are mostly on the bare glass, even with

a final Cu thickness of 10μm (Figure 8(c, d)). This is a

noteworthy result, considering that a previous study on

electroless deposited Cu adhesion to glass showed that only

glass with a surface roughness exceeding 500 nm was sufficient

Open viaGlass ABF

(a)

Open via

Glass ZF

(b)

A: Mask side B: Back side

Mask

Glass

PrimerPR

uv uv uvuv

A

BPrimerPR

1210

[3]. This can be attributed to some glass surface roughness

induced by excess laser energy used to ablate the primer.

Figure 8. Optical micrographs of fabricated daisy-chain

structures on (a) ABF and (b) ZF after CO2 laser ablation and

(c) ABF and (d) ZF after UV laser ablation.

In addition to the primer surface, the glass via sidewalls

were conformally plated with Cu without any observed

delamination or discontinuities (Figure 9). The glass-Cu

interface in the via sidewalls has no organic primer film; it is a

direct metal to glass interface. It is well known that in order for

metals to adhere directly to oxide glasses, there must exist

either rough surfaces for sufficient areas for secondary bonding

through Van der Waals forces and mechanical anchoring, or a

thermodynamically stable intermediate metal oxide which

provides primary bonds, or a combination of both [2, 5-7].

Since no interfacial oxides are present to form chemical bonds

from Cu to glass, the apparent adhesion must be primarily due

to a combination of secondary bonding and mechanical

anchoring. As previously hypothesized, surface roughness

induced from glass via formation processes provided the

necessary surface area for sufficient Van der Waals adhesion

and mechanical anchoring [1]. This is consistent with the

adhesion observed in the Cu pads on UV-processed samples:

an excess dosage of the UV pulse would have affected the glass

surface in a similar way as when glass vias are drilled by UV

laser, producing a surface that adheres well to electroless

deposited Cu.

Daisy-chain structures were designed having lengths

between 2 to 66 vias. Figure 10 shows the measured daisy-

chain resistance of representative samples with ABF and ZF

plotted against the daisy-chain length after pre-conditioning

and before thermal cycling. A least-squares linear fit was

applied to the data with a y-intercept of zero, and the R-square

values of 0.9845 and 0.9727 of ABF processed by CO2 and ZF

processed by UV laser, respectively, indicate good linearity and

negligible irregularities in Cu daisy-chain conductivity. The

typical resistance per via extracted by the linear fit was about

10 mOhms per via. At the time of writing, samples were

undergoing thermal cycling; full results will be presented at the

conference.

Figure 9. Optical (top) and SEM (bottom) micrographs

showing cross section of fabricated Cu daisy-chains on ZF

primer on glass opened by CO2 laser.

Figure 10. Plot of daisy-chain resistance as a function of

chain length.

Although two primer materials were used in this study, the

process can be used with any thin film material that can be

laminated on glass with high adhesion, is easily patternable,

and is ready for electroless or other metallization processes. In

this case, both primer film materials demonstrated

compatibility with all attempted opening processes. Between

plasma etching and laser ablation, samples processed through

the former method yielded lower quality results than the latter.

The process is also open to other primer opening methods that

have not been discussed in this study. For example, a

photosensitive primer could be used and easily patterned by

traditional photolithography steps. The successful

metallization and structure fabrication on all samples confirm

the adaptability of the process. Finally, the daisy-chain

resistance measurements confirm that functional structures

have been successfully fabricated.

Conclusions

A novel method to metallizing glass with vias has been

investigated and demonstrated for the first time. Glass with vias

has been laminated with various thin primer film materials

which were subsequently opened by different methods,

demonstrating the versatility of the process. Daisy-chain

structures in ultra-thin glass interposers fabricated by

1211

laminating ABF or ZF and opened by CO2 or UV laser showed

good electrical behavior. The investigated process presents new

options for metallizing glass interposers manufactured with

through vias.

Acknowledgments

The authors would like to acknowledge Vanessa Smet,

Fabian Benthaus, and Tim Fleck for assistance in processing as

well as Chris White and Jason Bishop for their guidance and

support in sample fabrication.

References

1. V. Sukumaran, T. Bandyopadhyay, V. Sundaram, and R.

Tummala, "Low-Cost Thin Glass Interposers as a Superior

Alternative to Silicon and Organic Interposers for

Packaging of 3-D ICs," Components, Packaging and

Manufacturing Technology, IEEE Transactions on, vol. 2,

pp. 1426-1433, 2012.

2. P. Benjamin and C. Weaver, "The Adhesion of Evaporated

Metal Films on Glass," Proceedings of the Royal Society of

London. Series A. Mathematical and Physical Sciences, vol.

261, pp. 516-531, 1961.

3. T. Huang, V. Sundaram, P. M. Raj, H. Sharma, and R.

Tummala, "Adhesion and reliability of direct Cu

metallization of through-package vias in glass interposers,"

in Electronic Components and Technology Conference

(ECTC), 2014 IEEE 64th, 2014, pp. 2266-2270.

4. P. Ogutu, E. Fey, P. Borgesen, and N. Dimitrov, "Hybrid

Method for Metallization of Glass Interposers," Journal of

The Electrochemical Society, vol. 160, pp. D3228-D3236,

2013.

5. A. P. Tomsia and J. A. Pask, "Chemical reactions and

adherence at glass/metal interfaces: an analysis," Dental

Materials, vol. 2, pp. 10-16, 1986.

6. Y. H. Kim, Y. S. Chaug, N. J. Chou, and J. Kim, "Adhesion

of titanium thin film to oxide substrates," Journal of

Vacuum Science & Technology A, vol. 5, pp. 2890-

2893, 1987.

7. N. Jiang and J. Silcox, "Observations of reaction zones at

chromium/oxide glass interfaces," Journal of Applied

Physics, vol. 87, pp. 3768-3776, 2000.

1212