Author Archives: sdavis

IFTLE 250 Toshiba Extends TSV Stacks to NAND; UMC to Supply TSV Based Silicon Interposers; Nvidia’s Pascal Coming in 2016 with HBM2

By Dr. Phil Garrou, Contributing Editor

Wow, the big 250!

Thanks to all of you that continue to read my updates and opinions. Thanks to all of those who have supplied me information to pass on to all of you.

I have told many of you that these blogs are my way of keeping myself up to date. I’ll keep doing these as long as I’m still having fun and as long as Extension Media and Pete Singer want me, too.

Before we get to some major 3DIC news, long time readers know I never miss a chance to update you on how my granddaughters are doing. Hannah (11) and Madeline (7) were in NC visiting a few weeks ago, so I will share this shot with you.



and now the news…

Toshiba Extends TSV Memory Stacking to NAND

IFTLE (Insights From the Leading Edge) which started in August of 2007 as Perspectives From the Leading Edge in the now defunct Semiconductor International has mainly focused on advanced packaging with emphasis on 3DIC since that was the hot and coming technology. By blog #12, I announced that Toshiba had would be commercializing CMOS mage Sensors with TSV (albeit one layer devices, but most would conclude, like I do, that this was the commercial start of what we now call 3DIC.

It is therefore somehow fitting that as I hit IFTLE 250 (and IFTLE + PFTLE 378 ) that Toshiba would be the first to announce the extension of TSV technology to NAND. [link] For years, I have been taking the position that NAND could not come till DRAM was commercialized since it is a less costly product. Well, since TSV DRAM has now been announced in many forms by the major DRAM players (Samsung, Hynix and Micron) I guess its now time for the NAND announcement.

Toshiba’s TSV technology achieves an I/O data rate of over 1Gbps which is higher than any other NAND flash memories with a low voltage supply (1.8V to the core circuits and 1.2V to the I/O circuits) and approximately 50% power reduction of write operations, read operations, and I/O data transfers.

They see this stacked NAND flash memory providing the ideal solution for low latency, high bandwidth solutions in flash storage applications, including high-end enterprise SSD.



Sold as 152 pin BGAs, the 8 stack modules are 14 x 18 1.35mm and the 16 stack devices are 14 x 18 x 1.90mm. 

UMC Enters HVM on SI Interposers for AMD FIJI

We announced the AMD FII graphics processor would use high bandwidth memory and silicon interposers in IFTLE 240 “AMD Introduces HBM on Fiji R9 390X GPU.”

WBM on interposer


It has recently been announced that the silicon interposers are being supplied by UMC making them the second entrant into the commercial silicon interposer business (after TSMC and their CoWoS process). The interposers are reportedly being manufactured by UMC in Singapore, at 300mm Fab 12i.

FIJI graphics module

AMD FIJI graphics module with 4 stacks of HBM

The GPU package, measuring 50 x 50 mm, will actually reduce the graphics card PCB size, because the memory has been moved to the GPU package, with four 1024-bit HBM1 stacks surrounding the GPU die (n the package see below). 1GB of HBM memory takes up 95 percent less than the same amount of GDDR5 memory, so you free up a lot of room on a circuit board. AMD has yet to announced whether their next gen Greeenland GPUs, which will use HBM2 memory stacks will be made on TSMC’s 16nm or Samsung’s 14nm process.

SK Hynix ships their HBM stacked to UMC who integrates the HBM stacks and GPU onto their silicon interposer for AMD.

Nvidia Pascal Coming in 2016

Not to be outdone by AMD, Nvidia has announced further details on the their Pascal graphics processor due out in 2016.

Nvidia roadmap

Nvidia’s Pascal GPU ( GP100) will feature a 4096 bit bus and four HBM2 stacks each up to 8 memory layers. The Pascal chip set will reportedly be manufactured on both TSMC’s 16nm FinFET process [link 1] and Samsungs 14nm FinFET process [link 2] later next year. Rumors are that Nvidia has taped out the GP100 at TSMC and thus should be expecting the first chips from them in Q2 2016. The GPU will be made with either 4 or 8 memory layers in the HBM stacks. They believe the Pascal GPU will be able to achieve 10x better performance compared to Maxwell.

Nvidia Pascal GPU Module

Nvidia Pascal GPU Module

OK, Intel, your turn…we’re waiting !

For all the latest on 3DIC and advanced packaging, stay linked to IFTLE…

IFTLE 249 Merger Mania or Simple Economics? ; Wide Band Gap Semiconductors and Advanced Cooling

Merger Mania or Economics 101?

SST has recently shared an article by IC Insights entitled “Tsunami of M&A deals underway in the semiconductor industry in 2015.” [link]

While IFTLE certainly agrees with their conclusion, that mergers are occurring on a larger scale and more rapidly than in the recent past, we do not agree with their characterization “…It would be hard to characterize the huge wave of semiconductor mergers and acquisitions occurring in 2015 as anything but M&A mania, or even madness.”

As IFTLE has been preaching for some time now Economics 101 tells us that this is the natural evolution of any industry. As was explained in IFTLE 241 “Simply Obeying the Laws of Economics” [link] all industries go through cycles or stages as they mature and most of our industry is now in late stage 3 or early stage 4. IBM saw what was going on, assessed their position and got out.

A major part of these later stages is consolidation (or M&A deals). Our industry has always gone through boom or bust cycles, the difference this time is we are further along the industry maturity pathway. Actually economics also tells us that as this massive consolidation occurs those that remain will actually see less of the boom-or-bust cycles because the industry as it matures will stabilize. Those that understand this understand that we are getting ever closer to the “acquire or be acquired” phase in the electronics business and they are making their moves. Those that don’t….well they won’t be around to much longer. Still confused? Try:



That said, IFTLE certainly agrees with the closing statement of the IC Insights article “…the increasing number of mergers and acquisitions, leading to fewer major IC manufacturers and suppliers, is one of major changes in the supply base that illustrates the maturing of the industry… The strong movement to the fab-lite business model, and the declining capex as a percent of sales ratio, all promise to dramatically reshape the semiconductor industry landscape over the next five years.”

Wide Band Gap Semiconductors Require Advanced Cooling Solutions

While much of our industry is maturing (as discussed above) new segments are always evolving. Wide bandgap semiconductors is such a new semiconductor segment.

Wide bandgap (WBG) semiconductor materials allow power electronic components to be smaller, faster, more reliable, and more efficient than their silicon based counterparts. These capabilities make it possible to reduce weight, volume, and life-cycle costs in a wide range of power applications. WBG semiconductors have great potential as enabling materials in high-density power applications, satellite communications, and high-frequency and high-power radar making them the darlings of the defense industry.

high bandgap

WBG materials have the potential to:

Reduce energy losses: Eliminate up to 90% of the power losses that currently occur during AC-to-DC and DC-to-AC conversion.

Higher-voltage operation: Handle voltages more than 10X higher than Si based devices, greatly enhancing performance in high-power applications.

Higher-temperature operation: Operate at temperatures over 300°C (twice the max temp of Si-based devices).

Higher frequency operation: Operates at frequencies at least 10 times higher than Si-based devices, making possible new applications, such as radio frequency (RF) amplifiers. [link]

WBG Thermal Challenge being addressed by DARPA

While WBG materials are rapidly gaining acceptance in numerous applications, one of the challenges that remains to be addressed to fully exploit the promise of WBG semiconductors is thermal management . Alternative cooling solutions will be required to withstand the high temperatures of WBG devices.

bar cohenDARPA has been addressing this issue since 2013 under their MTO ICECool program headed up by Dr. Avram Bar-Cohen. Bar-Cohen, recently recognized by IEEE with the IEEE CPMT field award (the highest level packaging award available in the IEEE) is regarded as one of the world experts in thermal management issues.

The ICECool program has many of the countries key aerospace companies such as BAE, Boeing, HRL, Lockheed Martin, Northrup Grumman and Raytheon, developing solutions for this WBG thermal problem.

The key vs previous thermal solutions (see figure below) is to remove heat at the chip as close to the transistor heat generation as possible.


Click to view full size.

Although I cannot share the specifics with you here (ITARS restrictions), I can say that significant advances are being made in several of the ICECool programs that should allow WBG semiconductors to make major impact on our microelectronics future.

For all the latest in 3DIC and other advanced packaging, stay linked to IFTLE…

IFTLE 248 ECTC 4 Oxidation of EMC in Thin Packages; Compression Molding Large Panels; TSV Noise Coupling in 3DIC

By Dr. Phil Garrou, Contributing Editor

Continuing our look at the 2015 ECTC.

Freescale – High Temperature Storage of Ultra-Thin Molded Array Packages

Ultra-thin molded array packages (TMAP) with package thickness ≤ 500 μm are desirable for applications where system integration space is limited. Such ultra-thin packages require careful selection of the epoxy molding compound (EMC) to control package level warpage.

In this presentation, the ultra-thin package (8 mm x 8mm) had an EMC thickness of 0.250mm, substrate thickness of 0.1mm. This package was found to exhibit tensile warpage of 160μm after 175°C, 500 hrs of high temp storage (HTS).

The primary mechanism for this warpage behavior was found to be thermal oxidation of the EMC in HTS. These conditions caused thermo-oxidative crosslinking leading to densification and shrinkage of the EMC inducing stresses leading to package warpage. Oxidation also changed the CTE and elastic modulus of the EMC making it more brittle and stiffer.

Cross-sectioned samples showed that the packages subjected to 175 °C, 500 hrs had a “dark brown” appearance closer to the package edges and progressing towards the die region (see below). This was not seen for the T0 samples. This dark brown area is due to EMC that has been oxidized under the high temperature exposure. Oxidation is proportional to time, temp and exposed surface area. Thus thinner packages show more of an impact of the same time/temp oxidation conditions than thicker packages since a higher % of the overall material present would be oxidized.

This shrinkage of the outer layer of EMC causes stress on the package which generates increased warpage of the package.

oxidized EMC

Fraunhoffer IZM & TU Berin – Compression molding for Large Area Panel Processing

The group at Fraunhoffer IZM and Tech Univ Berlin have studied large area (18 x 24”) compression molding of panalized packages. The large panel size was selected to achieve process compatibility with cost efficient PCB processes. They examined the so-called “mold first” process which starts with die assembly on an intermediate carrier followed by overmolding and debonding of the molded wafer/panel from the carrier. The redistribution layer based on e.g. thin film or PCB technology is finally applied on the reconfigured molded wafer/panel.

Using an ASM Siplace CA3 5680 chips were placed on the panel with an assembly speed of around 6500 chips/hr. An 8×8 mm, 2-chip package consisting of two 2×3 mm chips the test vehicle.

They used a APIC Yamada molding tool with a cavity size 24”x18” and thickness range from 100 – 5000μm. The fig below shows the principle of compression molding with mold cavity in the upper tooling and the reconfigured wafer in the lower tooling with liquid encapsulant.

comp mold

EMC Materials can be can be either liquid, granular or sheet compounds. Mold embedding materials should have low chemical shrinkage, low cure temperature and match thermo-mechanical properties for low warpage of the molded panel and low die shift after molding.

Both the liquid and granular material underwent a significant change in mechanical properties after post mold curing (PMC) since the curing reaction is NOT complete after leaving the molding machine. Their data indicates that heating and cooling the molded panel before PMC and carrier release should be avoided due to CTE mismatch and the related length changes between silicon dies, carrier material and moldings compounds. Such induced stresses causes a high risk for cracks in the molded panel especially when it is not fully cured.

They report that the most important requirement on the mold embedding process on large panels is the positional accuracy of the embedded dies after mold and cure.

IMEC – Noise Coupling between TSV and Active Devices

IMEC notes that TSVs can introduce noise coupling arising from electrical coupling between TSVs and the active devices. They investigated the TSV noise coupling to FinFETs and planar transistors up to 40 GHz.

By analyzing and comparing the impact of TSV noise on FinFET and planar device performance, the dominant coupling mechanisms were identified. For planar nMOSFETs, noise amplification through the bulk transconductance dominates for the ON state, leading to large noise coupling. For FinFETs the main coupling mechanism was capacitive even in the ON state. They conclude that nFinFETs have better noise coupling immunity than planar nNMOSFETs, i.e. TSV noise coupling to FinFET device is 20dB smaller than planar device at 1GHz and 4-8dB smaller at 40 GHz. They explain that the main reason for improved FinFET noise immunity is that the FinFETs are less sensitive to substrate bias due to the stronger gate control w.r.t. planar MOSFETs.

They have also investigated four different TSV architectures to predict their impacts on the noise coupling: 5um/50um via-middle TSVs with 200nm oxide liner; scaled 3um/50um via-middle TSVs with 200nm oxide liner; 5um/50um via-middle TSVs with 400nm oxide liner; and via-last TSV architecture with thick (3um) polymer liner (“Donut” TSV).

They conclude that scaling the TSV diameter, using thicker liner materials, and polymer based liner materials with smaller dielectric constant all reduce the coupling level mainly at low frequencies where TSV liner capacitance is the dominant factor on noise coupling. At 100 MHz, for the 3/50 um and “Donut” TSVs, the noise coupling reduces by about 4dB and 20dB respectively.

They conclude that In order to extract the KOZ accurately, coupling induced current variations must be considered together with the stress induced current change. Stress induced current change decreases rapidly with increasing distance; i.e. when the active device is 20 um away from TSV, the stress induced current change is close to zero for both planar and FinFET devices.

However the coupling induced current change decreases with distance much slower, which means the noise coupling can have significant influence on the KOZ. When an active device is located at 10um away from TSV, the current change induced by TSV stress is only 0.2% for FinFET device and 0.6% for a planar device, but these values increase to 2.23% and 3.55% when the impact from TSV noise coupling is added up at 10 GHz (2.03% for FinFET device and 2.95% for planar device). They conclude that it is important to use noise mitigation techniques such as “substrate contact and guarding” to reduce the electromagnetic coupling effects in order to minimize the KOZ.

For all the latest on 3DIC and advanced packaging, stay linked to IFTLE…

IFTLE 247 ECTC part 3: More Thermo Compression Bonding from Intel and ASM

By Dr. Phil Garrou, Contributing Editor

In IFTLE 245, we looked at some of the key thermo-compression bonding (TCB) papers at ECTC. Is there any question that TCB is real and will be the next big bonding technology ? This week, more coverage on this very important new assembly process from Intel and ASM.

Intel & ASM – TCB for fine pitch Flip chip (C2)

Intel introduced TCB into high volume manufacturing in 2014. As substrate and die become thinner and solder bump sizes and pitches get smaller, the thin organic substrate tends to warp at room temp and as the temp is increased during the reflow process. The thin die can also demonstrate temperature dependent warpage, which can come into play during the reflow process. The extent of warpage of the substrate and die at high temperatures can overcome the natural solder surface tension force leading to die misalignment with respect to the substrate, resulting in tilt, non-contact opens (NCO) and in some cases solder ball bridging (SBB). The figure below shows shows these various defects.

C2 defects

In the Intel TCB process, the substrate with pre-applied flux is held flat on the hot pedestal under vacuum. The die is picked up by the bond head, held securely and flat on the bond head with vacuum. After the die is aligned with the substrate, the bond head comes down and stops when the die touches the substrate. A constant force is then applied while the die is heated up quickly beyond the solidus temperature. As soon as the solder joint melts, the die is moved further down (solder chase) to ensure all solder joints are in contact. The die is held in position allowing the solder to reflow completely, and to wet the bump pads and copper pillars. While the solder is still in the molten state, the bond head retracts upwards controlling the solder joint height. The bond head then releases the vacuum holding the die and moves away as the solder joints have solidified. The major process parameters, i.e temperature, force and displacement are continuously monitored during the TCB bonding process.

A schematic of the bonding tool is shown below.


Large differences in the CTE between the organic substrate and die results in different magnitude of expansions when heated which can lead to serious bump offset at corners. To minimize the thermal expansion mismatch, the substrate is processed at a lower temperature (e.g. 140ºC) while the die and solder is rapidly heated up for reflow and cooled down for solidification using a pulse heater with heating ramp rate exceeding 100ºC/s and cooling ramp rate exceeding 50C/s. This reduces the heat transfer to the substrate. The bulk of the substrate can remain at low temperature and does not expand extensively.

ASM – High Throughput Thermal Compression Bonding

In another ASM paper on TCB, they examined what they call liquid phase contact (LPC) TCB. The goal is to increase the throughput of the TCB process. Process flow is shown below. Flux is printed or sprayed on the substrate. Then the bonding head picks up a die from the carrier at an elevated temperature, but below the solder melting point. Hen the bonding head is heated up to a temperature higher than the solder melting point and the chip is aligned with the substrate. The chip is then contacted and wetted on the substrate at a predetermined bonding height. After a predetermined bonding time, the bonding head can move is cooled down to a temperature below the melting point of solder.



They claim this results in attachment of 1200 units/hr vs 600 for the std. TCB flux process.

For all the latest on 3DIC and advanced packaging, stay linked to IFTLE…


IFTLE 246 ECTC 2: IMEC 2.5/3D Process Developments and Low Temp Bonding for Si on Diamond

By Dr. Phil Garrou, Contributing Editor

Continuing our look at the 2015 ECTC:

Kaiung Paik – KAIST and Jay Im – U Texas

Kaiung Paik – KAIST and Jay Im – U Texas



Nancy Stoffel – GE, Chris Bower & Carl Heinz Bock –                                                                                        Emerging Tech Committee Chairs & Marsha Tickman -                                                                                       Exec Dir CPMT

Nancy Stoffel – GE, Chris Bower & Carl Heinz Bock –
Emerging Tech Committee Chairs & Marsha Tickman –
Exec Dir CPMT

Stress & Bowing in Passive Silicon Interposers – IMEC

Three modifications of the structure of a 4 BEOL layers with 10×100μm TSV Si interposer are proposed to mitigate the tensile stress and release the interposer warpage. The use of a thicker and more compressive PMD layer enable a wafer bowing reduction of 75% after TSV processing. By reducing the thickness of Metal1 (ground plane with 75% Cu density) from 1000 nm to 300 nm, the single contribution of the Metal1 module to wafer bowing was reduced by 37%. Finally, by using a more compressive oxide (from -170 MPa to -230 MPa) to process the Via2/Metal3/Via3/Metal4 layers (total of 8μm thickness), their contribution can change from tensile to compressive, inducing a total reduction of -120%. In total, a global bowing reduction after full front side processing of 75% was measured compared to the initial interposer structure.



Effects of Packaging on Mechanical Stresses in 3DIC – IMEC

IMEC has studied the mechanical stress induced in 3D stacks by different packaging process steps. The 3D stacks used in this work are assembled using two identical dies containing a number of stress sensors which are designed and manufactured in 65nm technology. It is observed that the contribution of the package substrate and the die-attach process to the redistribution of mechanical stress inside the 3D stacked IC is more significant than the one of the EMC and that the influence of packaging on the shape and amplitude of local stress around the micro-bumps is not significant. These observations are supported by the measurements of stress done using micro-Raman spectroscopy and are correlated with the results of finite element modeling and with optical warpage measurements of different packaging configurations.

Advanced Metallization Scheme for 3 x 50 um TSV Middle Process – IMEC

Scaling down the TSV diameter from 5μm to 3μm is very attractive for the 3D IC implementation in more advanced CMOS nodes. For instance, stress caused by the mismatch between the coefficient of thermal extension (CTE) of Si and Cu may generate strain in the Si around the TSV, degrading the device performance of transistors located close to a TSV. To reduce the impact on transistors, a so called keep-out-zone (KOZ), is generally defined around the 3D TSVs. This keepout-zone is however significantly smaller when scaling down the TSV diameter from 5μm to 3μm. When increasing the aspect ratio of the TSV from 10:1 to 17:1 (for 3μm diameter and 50μm depth), the conventional PVD barrier and seed options reach their conformality limits. Very thick barrier/seed layers need to be deposited in order to assure a continuous film at the bottom of the TSV. This not only fundamentally limits the extendibility of this integration scheme, but also increases the PVD deposition cost itself and the required CMP time, among other technical challenges. For these reasons, a new advanced and scalable TSV metallization scheme was developed.

Atomic layer deposition (ALD) has emerged as a key enabling technology for conformal film applications such as TSV oxide liner. Typical step coverage or conformality of other CVD oxide films is only 60-75% for high aspect ratio TSVs . In contrast, the VECTOR ALD oxide film shows 100% conformality.

ALD WN serves as a barrier layer and is deposited on the Altus Max tool. It is highly conformal with >90% step coverage regardless of geometry. ALD WN is deposited at 375°C and has excellent adhesion to ALD oxide and subsequent ELD NiB. The conformality of the ALD process results in a pinhole-free WN layer, as opposed to barriers deposited by PVD, which potentially struggle with pinholes at the bottom of high aspect ratio TSV.

NiB electroless deposition on WN barrier was carried out on a Lam ELD2300 tool using a plating chemistry developed at Lam. The entire deposition process was made of several sequential steps such as a brief pre-clean, activation, deposition, rinse and dry. The concentration of reducing agent and nickel ions as well as the pH and temperature are controlled to maintain optimum deposition condition for seed formation during the NiB deposition. After deposition of ELD NiB, TSV copper electrofill is processed on a Lam SABRE 3D electroplating system using an industry standard acid copper sulfate electrolyte with a Lam

exclusive organic additive package. The conductivity and corrosion resistance of the Lam ELD NiB film enable compatibility with the electrofill process on 300mm wafer scale. Bottom-up fill of the vias proceeds without the need for the additional copper seed film that has been used with other conformal metal liners (e.g. Co or Ru).

Because of the high conformality of liner, barrier and seed layers, this proposed Via-middle metallization scheme is believed to be scalable to even higher aspect ratio TSVs with 2μm diameter.

Silicon on Diamond by Low Temp Bonding – Kyushu Institute of Technology

Diamond has long been known as an excellent thermal material, but has been a hard material to process. Researchers at Kyushu Univ had recently developed a low temp diamond bonding technique. The process flow is shown in the figure below. The SOD fabrication process includes: (a) Fabrication of thermally grown SiO2 film on Si wafer as a bonding substrate. (b) Deposition nanocrystal diamond film on Si wafer as another bonding substrate. (c) Removal of big nanocrystal diamond particles. (d) Deposition of SiO2 film on the nanocrystal diamond film by CVD. (e) CMP of the SiO2 film. (f) Thinning of the CMP SiO2 film by 2.5% HF solution. (g) Surface chemical cleaning of the bonding substrates by piranha solution. (h) Surface activation by O2 plasma. (i) Bonding of the substrates.

Si on Diamond


The middle figure shows (a) the morphology of 300nm-thick of nanocrystal diamond on the wafer 525nm-thick wafer. The surface roughness of nanocrystal diamond was characterized by AFM which was 13.9nm rms approximately. Generally, the roughness of thin film from the CVD process is dependent on the substrate. Thus a surface roughness of SiO2 that we plan to deposit in the next step is probably close to 13.9nm rms.

In addition, the big nanocrystal diamond particle may be an obstruction to the thinning SiO2 before bonding. Therefore, we remove the big particles using CMP equipment before deposition of SiO2 film on it. Fig. (b) shows the morphology of CVD-SiO2 500nm thick deposited on substrate after remove big particle already. The roughness was reduced to 8.1nm rms compared with the substrate. Fig (c) shows the morphology of CVD-SiO2 after polishing by CMP. The roughness was reduced to 0.50nm rms.

An average CVD-SiO2 roughness of > 1 nm rms failed to bond, the bonding results show 95% confidence level for bonding with a roughness is 0.97±0.03nm rms.

The figure on the right shows the cross section of the stack shows the dimensions of the overall structure.

For all he latest in 3DIC and advanced packaging, stay linked to IFTLE…


IFTLE 245 2015 IEEE ECTC part 1 Thermo compression Bonding

By Dr. Phil Garrou, Contributing Editor

Over the next few weeks of the summer, I will be covering as much of the 65th IEEE ECTC as I can. They call themselves the “Premier International Packaging Conference” and they are. Authors from companies, research institutions, and universities from over 25 countries presented their work at ECTC, illustrating the conference’s global focus. In addition, the ECTC offered 16 Professional Development Courses (PDCs) and Technology Exhibits.

This year there were 350 papers presented in 36 oral sessions covering 3D and TSV technologies, wafer level packaging, electrical and mechanical modeling, RF packaging, system design, and optical interconnects.

This years conference leaders include (from L to R) Sam Karikalin, Broadcom, Beth Kesser, Qualcomm, Alan Huffman, RTI Int and Henning Braunisch, Intel.

ectc 1


First, let’s look at some advances in thermos-compression bonding.


Jie Fu of Qualcomm discussed “Thermal Compression Bonding for Fine Pitch Solder Interconnects”. Mass reflow-based interconnects, using either solder bump or Cu-column on bond on lead are the typical low-cost flip chip assembly approachs used by industry. These interconnects face challenges related to shorting and non-wets at sub 100um pitches. Transitioning below 100um pitch requires a new approach, such as thermos-compression flip chip (TCFC). While TCFC provides higher accuracy bonding and allows for use of smaller solder cap which enables tighter FC pitch, it also presents new challenges. The major challenges for TCFC bonding include lower throughput and control of non conductive paste (NCP) voids. Overall, bond head ramp rate, temperature uniformity, peak temperature and dwell time must be fine tuned in tandem to compensate for manufacturing tolerances and to get the desired end of line solder joint structure. In addition, controlling the temp exposure for the NCP material before NCP cure is critical to enable a robust TCFC solder joint. To much thermal exposure and the NCP begins to cure prior to solder melting, which can leading to NCP entrapment and unreliable TCFC solder joints. Laminate surface finish is also an important variable.


In a similar study Cho and co-workers at GlobalFoundries presented “Chip Package Interaction Analysis for 20-nm Technology with Thermo-Compression Bonding with Non-Conductive Paste”. Strong market demand for finer pitch interconnects to enable higher I/O counts in a smaller form factor is driving another transition from conventional MR bonding process to thermo-compression bonding using non-conductive paste (TC-NCP). FEA simulation results for TC-NCP vs mass reflow show that TCNCP has significantly reduced thermo-mechanical stress at the ULK level and the bump level.


Horst Clauberg of K&S discussed “High Productivity Thermocompression Flip Chip Bonding”. There is tremendous effort by IDMs, OSATs, materials suppliers and equipment suppliers to bring thermos-compression bonding to commercial reality. The most significant technical challenges have for the most part been solved and limited commercial production is taking place. However, relatively low throughput and high equipment cost create adoption resistance, especially in the all-important consumer market.

Due to the relatively high cost, the only component of the industry clearly adopting thermocompression bonding is the advanced memory segment, such has hybrid memory cube (HMC) and high bandwidth memory (HBM). The rate of adoption for applications processors, GPUs and the like may depend on the rate at which throughput and cost can be improved.

Thermocompression bonding can be segmented into two different processes. The first process differentiation is whether the underfill is pre-applied before the semiconductor chip is mounted or not. Pre-applied underfill comes either as a film applied to the die or as a paste applied to the substrate. In both cases the underfill must not only create a void-free bond, but also provide flux to remove oxide on the solder caps. The alternative process is thermocompression – capillary underfill (TC-CUF) where the die is underfilled in the same way as std flip chip,except that the underfill process is much more challenging because of the more narrow bondline of a typical thermocompression bonded device. In TC-CUF, flux can be applied either by dipping the die into flux before bonding, or applying flux to the substrate.

A K&S cost-benefit analysis of a C2 (copper pillar bump) TC bonding process was used to look at the total packaging cost. Besides enabling higher I/O counts and finer pitch interconnections through better control of the stress and warpage, the actual bonding process is just a small contribution to the overall assembly cost. K&S shows that the incremental assembly cost adder for thermocompression bonding is actually rather small in a high UPH TC bonder. The hurdle to wide-spread adoption of the TC bonding is more likely the initial capital expenditure associated with buying new equipment when depreciated infrastructure already exists for mass reflow processes. Adoption of the technology will therefore be driven by technical need and market forces. TC bonding will enable higher I/O counts and finer pitch interconnections than traditional interconnect methods through better control of the stress and warpage between devices and the substrate. Once the infrastructure is established, they predict that the cost will decrease directly proportional to throughput and they have demonstrated that throughputs of 1000uph are possible.

Fig 2

Amkor / Qualcomm

Doug Hiner in a joint presentation between Qualcomm and Amkor presented “Multi-Die Chip on Wafer Thermo-Compression Bonding Using Non-Conductive Film”. Non-conductive films have been in development as a replacement to the liquid preapplied underfill materials used in fine pitch copper pillar assembly.

Several assembly methods are available for chip on wafer assembly including: (1) traditional chip attach with mass reflow (MR) and capillary underfill (CUF), (2) thermo-compression

bonding (TCB) of copper pillar interconnects using nonconductive paste (NCP) underfill (TCB+NCP), and thermocompression bonding of copper pillar with non-conductive film (NCF) underfill (TCB+NCF).

The TCB+NCP process carries concerns with the underfill time on stage which prevents the dispensing of the NCP material across the wafer prior to the chip bonding process. This constraint effects process costs significantly. The TCB+NCF process to date have not met the cost/benefit needs of the industry. NCF assembly provides significant improvements in the design rules associated with die to package edge, die to die, and fillet size. The NCF process also resolves the time on stage concerns associated with the NCP process by laminating the NCF material to the bonded die instead of to the interposer or receiving wafer surface.

Development has proven the feasibility of a multi-die (gang) bond chip on wafer assembly process. Key assembly steps have been validated and major issues have been mitigated through optimization of materials and process parameters. A scale up phase of development has been initiated which targets the bonding of 8 die (4 units) in a chip on wafer format. Assembly cost of ownership estimates (OSAT) suggest cost parity between 8-die gang bonding and traditional mass reflow…

For all the latest on 3DIC and advanced packaging, stay linked to IFTLE.

IFTLE 244 3D Stacked CMOS Image Sensors; IEEE 3DIC Conf

By Dr. Phil Garrou, Contributing Editor

At the recent 2015 Int Image Sensor Workshop, Ray Fontaine of Chipworks presented a review of “The State-of-the-art of Mainstream CMOS Image Sensors” Chipwork’s estimate, from other market research firms, is that the CIS market in 2014 was ~ $9B. Of this total it is estimated that Sony, Samsung and Omnivision hold > 66% market share driven by mobile phone and tablet camera chips.

Their look at the patent literature shows the field continues to grow with 2500 patents filed in 2014, the majority of them being processing patents.

cmos image sensor patents

Stacked Chip CIS

3D stacked CIS became a reality in 2012 when Sony announced the worlds first stacked chip CIS in consumer cameras. In 2013 they introduced the 8 MP ISX014 in a tablet computer [ref]

[P. Jagodzinski, “Sony ISX014 ¼ inch 8 MP 1.12um pixel size Examor RS stacked back illuminated CIS imager process review” Chipworks March 2013 ]

The first gen chips employed via last TSV to connect pads on the Sony 90nm CIS die to the pads on the Sony 65nm ISP. The die stack was partitioned such that most of the functionality of a conventional system-on-chip (SoC) CIS was implemented on the ISP die; the CIS die retained the active pixel array, final stage of the row drivers, and comparator portion of the column-parallel ADCs. The CIS (left) and ISP die (right) are shown below.

Click to view full size.

Click to view full size.

Sony’s 13 MP IMX214 second generation stacked CIS chips were fabricated using its 90/65 nm (CIS/ISP) technology generation. The key work on the second generation stacked process was to use the CIS silicon only as the active pixel array substrate and move the column readout chain and peripheral transistors to the underlying ISP die.

In 2014, Sony announced they were using TSMC as a foundry for the 40nm ISP wafers on the Apple iPhone 6/6 Plus iSight cameras. These chips incorporate Sony 90 nm CIS wafers and TSMC 40 nm ISP wafers.

In 2015, Samsung and OmniVision have both been sampling small-pixel, stacked chip CIS.

Given the continued, aggressive stacked CIS development underway from independent device manufacturers (IDM) and foundries it’s predictable that stacked chip adoption will occur very rapidly over the next few years.

IEEE 3DIC Conference Sendi Japan Aug 31st

Click to view full size.

Click to view full size.

The IEEE International 3D System Integration Conference (3DIC) will be held in Sendai, Japan August 31-Sepember 2, 2015. After the first conference in San Francisco in 2009, the 2nd IEEE 3DIC Conference was held in Munich in 2010, and then Osaka in 2012. The forth conference was back in San Francisco in 2013 and the fifth conference in Cork, Ireland in 2014.

IEEE 3DIC 2015 will cover all 3D integration topics, including 3D process technology, materials, equipment, circuits technology, design methodology and applications. The conference invites authors and attendees to submit and interact with 3D researchers from all around the world. Papers are solicited in subject topics, including, but not limited to:

  • 3D IC Process Technology
  • 3D IC Circuits Technology
  • 3D Applications
  • 3D Design Methodology

For all the latest in 3DIC and advanced packaging, stay linked to IFTLE…

IFTLE 243 Amkor Fan Out Package Platforms

By Dr. Phil Garrou, Contributing Editor

Amkor recently held a customer Symposium covering their activities. Let’s take a look at some of the interesting points that they covered on their fan out package platforms.

2014 Amkor revenue is clearly dominated by the communications segment.

Amkor 1


When looking in general at the evolution packaging they see the largest focus on filling the gap between 1 and 10um as shown below.

Amkor 2


Their packaging roadmap to address this gap area is shown below and is tied to their Swift and Slim product families. Such ultra thin packages will have to be handled by temporary bonding to a rigid substrate in order to process them.

amkor 3


From a mobile products standpoint they see the 5+um range filling most of the needed requirements for density/IO, whereas the SWIFT product line will be needed for < 5um BB module (SiP) requirements.

Amkor 4


The SWIFT and SLIM processes are depicted below. SWIFT interconnect is carried out on an RDL bumping line by Amkor whereas 2-5um SLIM interconnect is fabricated by foundry.

Amkor 5


Since interconnect is fabricated first, higher densities can be achieved, i.e. 2-8um L/S for SWIFT vs           8-15um for traditional chips first FOWLP.

Amkor 6


In summary depending on the requirements of the application, different technologies are available and/or are being developed to meet those requirements.

Click to view full size.

Click to view full size.


For all the latest in 3DIC and advanced packaging, stay linked to IFTLE…


IFTLE 242 Advanced Packaging at the ConFab

By Dr. Phil Garrou, Contributing Editor

At the recent ConFab meeting in Las Vegas, aside from all the talk about consolidation (see IFTLE 241), Bill Chen from ASE and Li Li from Cisco put together a great Advanced Packaging session.

(L to r) Bill Chen (ASE), Ram Viswanath (Intel), Kevin Tran (Hynix), CP Hung (ASE), John Knickerbocker (IBM) and Li Li (Cisco)

(L to r) Bill Chen (ASE), Ram Viswanath (Intel), Kevin Tran (Hynix), CP Hung (ASE), John Knickerbocker (IBM) and Li Li (Cisco)

CP Hung, VP of R&D for ASE discussed “Integrating 3D IC into the IC Packaging DNA” Hung proposed that 2.5D IC significantly extends FCBGA technology as shown in the fig below.

Fig 2


Kevin Tran of Hynix announced that HBM (high bandwidth memory) has completed the qualification for mass production in March 2015. Each application has different memory requirements, but most common are high bandwidth and density. He indicated that packaging technology has become a key enabler for high performance, small form factor, low cost memory solutions.

Fig 3


Hynix is readying HBM 2 which will be applied for HPC, graphics, servers and network computing. High end graphics products have already been announced like Pascal at Nvidia and Greenland at AMD. Can Intel be far behind? IFTLE thinks not.

Fig 4


Ram Viswanath of Intel pointed out that “…the ability to monolithically integrate diverse functionality on the die has become impractical due to technology complexity and affordability” and that “on package integration is playing a key role in bringing diverse functionality into smaller form factor.” Key focus is on delivering

– performance for servers

– form factor for wearable products

– cost/form factor for client products

-low cost for future IoT products

Intel’s evolution of dense interconnect is shown below. The Xenon Phi for HPC uses memory stacks on an interposer (reportedly Micron HMC).

Fig 5


Intel compares  side-by-side multichip packaging to 2.5D interposers to 3D stacking in the table below. (note – IFTLE cannot support some of the conclusions on EMIB without seeing the actual data first).


Multichip                  side-by-side

2.5D Interposers

3D stack

Si interposer


IO/mm/layer 30-50 180-250 180-250 NA
IO/mm2 85-120 330-625 330 625
Elect perform (IO)        
Elect Perform (Power)        
Perform (Watt)        
Manuf complex (Yield)        
Thermal limits        

*2.5D designs are comparable

For all the latest in 3DIC and advanced packaging, stay linked to IFTLE…

IFTLE 241 Simply Obeying the Laws of Economics

By Dr. Phil Garrou, Contributing Editor

Following up on the recent blog by my comrade, Dick James.

There are laws and then there are laws. “Moore’s Law” to me is more of an observation. Gordon Moore simply noticed what was going on and commented on it.  Powerful laws, to me, are usually laws of physics like Newtons law of gravity or Einsteins Law of relativity.

When we consider the laws of economics, many economists would contend that consolidation is a law,   i.e. a natural process which has happened consistently to all industries since the industrial revolution. Through consolidation a mature industry usually has only a few (2-3) players ( for instance Boeing and Airbus in the aircraft manufacturing business) whereas young industries like the internet may initially have hundreds.

CarvilleSo what does this have to do with microelectronics you might say. Well, just ask the former employees of Altera and Broadcom. If this were the early 1990s James Carville (Clintons spinmeister) would respond “It’s consolidation stupid” [I was not a Carvile fan as you can tell by the picture I picked out !] Consolidation is what happens as industries mature. We are in the midst of it, it’s natural and probably unstoppable.

Lets first take a look at what’s happened in the hard disk drive segment of our industry.

Greater than 200 companies have been in the hard disk drive business since the 1960s. They initially competed on data density and latency and smaller form factors. Most of that industry has vanished through bankruptcy, mergers and acquisitions. Surviving manufacturers are Seagate, Toshiba and Western Digital. Seagate  acquired Samsung’s HDD business in 2011;  Western Digital (WD) merged with Hitachi’s HDD business in 2011. This gave Seagate 40% of the HDD market and WD ~ 48%. The remaining ~ 12% was owned by Toshiba who acquired Fujitsu’s HDD business in 2009. Thus by 2012 what was several hundred players had been whittled down to 3 by, I contend, the laws of economics.


Now let’s look at DRAM.


In 1980 there were 40+ DRAM fabricators but by 2015 we are down to Samsung, Hynix and Micron. See the trend?

The best description of whats happening, that I have seen is the 2002 Harvard Business Review article “The Consolidation Curve” by G K Deans et. al. Their main point is that all industries have similar life cycles and knowing where your company stands in the process can help you plot a winning strategy.

They divide up the stages of all industries as follows:

Stage 1: the combined market share of the three largest companies is between 10% and 30%. Companies in stage 1 industries aggressively defend their first-in advantage by building scale, creating a global footprint and establishing barriers to entry, i.e. protecting proprietary technology or ideas. Stage 1 companies focus more on revenue than profit, working to amass market share.

Stage 2: Stage 2 is all about scaling. Major players begin to emerge and buy up competitors.  The top three players in a stage 2 industry will own 15% – 45% of their market, as the industry consolidates. The companies that reach stage 3 must be among the first players in the industry to capture the most important markets and expand their global reach.

Stage 3: companies focus on expanding  core business and continuing to aggressively outgrow the competition. The top three industry players will control between 35% and 70% of the market with five to 12 major players remaining. This is a period of large-scale consolidation plays. Companies in stage 3 industries focus on profitability, and pare weak businesses units. The well entrenched in this phase will attack underperformers. Recognizing start-up competitors early on allows market leaders to decide whether to crush or acquire them. Stage 3 companies should also identify other major players that will likely survive into the next, and final, stage and avoid all-out assaults on them which could leave both players injured.

Stage 4: In stage 4 the top three companies claim as much as 70% to 90% of the market. Large companies may form alliances with their peers because growth is now more challenging. Companies in stage 4 must defend their leading positions. They must be alert to the danger of being lulled into complacency by their own dominance. Stage 4 companies must create growth by spinning off new businesses or buying into aligned fields to broaden their market presence.

When you understand this then headlines like the recent “The next three chip firms to be acquired: Atmel, Lattice and Cavium are the top take out candidates for the rest of 2015”[link]

As most of the segments of our industry enter late stage 3 or early stage 4 the only question is whether you will acquire or be acquired, or as Carville said “ It’s the economy stupid!”

For all the latest on 3DIC and advanced packaging, stay linked to IFTLE…