S.W. Yoon
STATS ChipPAC, JCET Group
The WLCSP was introduced in 1998 as a semiconductor package wherein all packaging operations were done in wafer form [1]. The resultant package has dielectric layers, thin film metals, and solder balls directly on the surface of the die with no additional packaging. The basic structure of the WLCSP has an active surface with polymer coatings and solder balls with bare silicon exposed on the remaining sides and back of the die. The WLCSP is the smallest possible package size since the final package is no larger than the required circuit area. Based on the small form factor and low cost, the number of WLCSP used in semiconductor packaging has experienced significant growth since its introduction. The growth has been driven aggressively by mobile consumer products because of the small form factor and high performance required in the package design. Although WLCSP is now a widely accepted package option, the initial acceptance of WLCSP was limited by concerns with the surface‐mount technology (SMT) assembly process and the fragile nature of the exposed silicon inherent in the package design. Assembly skills and methods have improved since the introduction of the package; however damage to the exposed silicon remains a concern. This is particularly true for advanced node products with fragile extreme low dielectric constant (low‐ELK) dielectric layers used in the silicon processing. One method commonly used to improve die strength and reduce silicon chipping during assembly is lamination of an epoxy film on the back of the die. The film is laminated and cured on the back of the wafer prior to singulation to strengthen the die, in spite of the fact it adds cost to the package. By the nature of the backside lamination process, the uncoated sides of the die continue to be exposed after dicing the wafer, and the silicon continues to be at risk for chipping, cracking, and other handling damage during the assembly process.
Despite the indisputable benefits of WLCSP, there are a number of concerns that have continued to plague the adopters of this technology since its inception:
Enter the eWLCSP that is a simple variation of the broader fan‐out wafer‐level packaging (FO‐WLP) platform (trade named eWLB for embedded wafer‐level ball grid array). eWLCSP retains the compelling benefits of WLCSP packaging while addressing many of the key concerns mentioned above. The structure of the package, the fabrication process, the unique advantages, and the preliminary product/reliability assessment are discussed here.
eWLCSP has been developed to provide five‐sided protection for the exposed silicon in a WLCSP. The process starts with an existing high volume manufacturing flow developed for eWLB fan‐out products. The implementation of this process flow into 300 mm diameter reconstituted wafers has been described in detail in previous presentations [2]. In this manufacturing method the wafer is diced at the start of the process and then reconstituted into a standardized wafer (or panel) shape for the subsequent process steps. The basic process flow for creating the reconstituted wafer is shown in Figure 22.1:
The eWLB process is unique since the reconstituted wafer does not require a carrier during the subsequent wafer‐level packaging processes.
After the reconstitution process, the reconstituted wafer is processed with conventional wafer‐level packaging techniques for the application and patterning of dielectric layers, thin film metals for redistribution and under‐bump metal, and solder bumps. In the final dicing operation, a thin layer of mold compound, typically less than 70 μm, is left on the side of the die as a protective layer. The back of the die is also protected with mold compound, although with a greater thickness than the sidewall layer. A schematic drawing of a typical structure is shown in Figure 22.2 for greater clarity. Alternatively, the backside mold compound can be removed, and the body is made thinner with an optional backgrind operation without damaging the protective sidewall layer. The remaining sidewall coating will continue to protect the fragile silicon sides of the die during the assembly operation.
The FO‐WLP process has been discussed in many venues, and it is recognized as an industry standard process. In the FO‐WLP process the area of the package is increased to allow for placement of redistribution layers (RDL) and solder balls outside of the silicon die area [3]. This packaging method allows the die to shrink to a minimum size independent of the required area for an array of solder balls at industry standard BGA ball pitches [4]. It also allows for novel multi‐die structures, 2.5D structures, and 3D structures. The FO‐WLP process has been qualified to a 28 nm process node with the same dielectrics and Cu plating as are used in the eWLCSP process described here [5]. The eWLCSP process data presented in this paper was generated with a 300 mm round reconstituted panel [2]. In the case of conventional FO‐WLP, the die is typically widely spaced to allow for the expanded RDL and bump area and the conventional saw street. In the case of eWLCSP, the die is closely spaced allowing for only the sidewall thickness in addition to a street area of 80 μm. The die size used in this evaluation was 4.5 mm × 4.5 mm similar to the construction shown in Figure 22.2. The final structure had two layers of polymer dielectric and one layer of plated Cu RDL with the solder ball mounted directly on the RDL without the use of a separate under‐bump metallurgy (UBM) layer. The process flow used is shown in Figure 22.3 and the details of the structure are shown in Table 22.1 [6, 7].
Table 22.1 Layer thicknesses.
PSV 1 (μm) | 7.0–11.0 |
RDL 1 (μm) | 7.0–10.0 |
PSV 2 (μm) | 7.0–11.0 |
Ball pitch (mm) | 0.4 |
Ball size (μm) | 250 |
Solder alloy | SAC405 |
Intuitively the process flow shown in Figure 22.3 would have higher cost since there are additional steps required for reconstitution at the start of the flow. There are two key factors that offset the cost of the additional steps required for the reconstitution to make this a commercially viable process. (i) Panel size scaling reduces the unit cost if the source silicon wafer is smaller than the reconstituted panel size. In the case of the 300 mm reconstituted panel used here, the cost is very competitive for silicon wafers with a diameter of 200 mm and below. The cost of processing a 300 mm reconstituted panel for WLCSP is approximately 1.7× the cost of processing a conventional 200 mm silicon wafer in WLCSP; however the units processed per panel increase by a factor of 2.3×, effectively offsetting the cost of reconstitution. (ii) Since known good die can be selected at the start of the process, advanced devices that have a lower electrical yield can be tested in wafer form prior to the process. If the incoming wafer has a probe yield of 85%, then 15% more units per reconstituted panel can be processed to offset the cost of the reconstitution process. Since the reconstituted panel size is no longer linked to the incoming silicon wafer size, the panel size can be increased over time and change from a round to a much larger rectangular format. This scaling to a larger panel size will provide a compelling cost reduction when compared with conventional WLCSP packaging methods where the round silicon format is maintained throughout the wafer‐level packaging process.
One difference in processing panels in the reconstitution flow is found in the attributes of the polymers that are used. In conventional WLCSP either polyimide (PI) or polybenzobisoxazole (PBO) is used as the dielectrics for planarization, stress buffering, and RDL insulation. In the case of the reconstituted panel, the mold compound has a lower temperature threshold than silicon, and sustained temperatures over 200 °C can cause degradation of the material. PI typically has a cure temperature of 380 °C, and PBO has a typical cure temperature of 300 °C and therefore cannot be used in the process. A new low temperature polymer has been developed for this application that has a cure temperature compatible with the 200 °C threshold temperature of the mold compound.
A scanning electron microscope (SEM) image of a cross section of an eWLCSP™ part created in the process is shown in Figure 22.4. In this case, a thicker sidewall protection layer was used to demonstrate the process on an existing production device running in the conventional 200 mm WLCSP production line. The device demonstrated equivalent electrical yield, component‐level reliability (CLR), and board‐level reliability (BLR) performance.
A second SEM cross section is shown in Figure 22.5 showing the finished package with a thin protective sidewall coating and the use of the optional backgrind to thin the body thickness.
A WLCSP product that is currently in production using a conventional WLCSP process can be converted to a eWLCSP product without any design change required, regardless of the current silicon wafer diameter. If a reduced thickness is required for the specific application, an optional backgrind step can be added to the process flow to reduce the body thickness while retaining the protective sidewall coating. Since the die is singulated at the start of the process, the manufacturing equipment and bill of materials are the same for any incoming wafer size. The initial backgrind and dicing tools are the only wafer size dedicated equipment required for the process. Very little process development and very little additional capital will be required to package 450 mm silicon wafers as eWLCSP.
Intuitively, eWLCSP would seem to have a higher cost over conventional WLCSP since there are additional steps required for reconstitution at the start of the FlexLine manufacturing flow. There are key factors, however, that offset the cost of the additional steps required for the reconstitution to make this a commercially viable process:
The unique attribute of the eWLCSP package is the protective sidewall coating. The protective layer is durable and will prevent silicon chipping on the side of the package. This protective layer has the ability to protect the silicon during socket insertion for test. This has been demonstrated through multiple insertion tests on completed products with no observed damage to the protective coating.
Robust reliability of 4.5 mm × 4.5 mm eWLCSP was reported with CLR and BLR tests. The eWLCSP process has passed standard reliability tests used in wafer‐level packaging including CLR, temperature cycle on board (TCoB), and drop test.
CLR was completed with the test conditions shown in Table 22.2. The evaluation results were confirmed by visual inspection and electrical test. No delamination of the protective coating was detected during the CLR evaluation.
Table 22.2 Component‐level reliability test results of eWLCSP.
Component‐level test | Condition | Status | |
MSL1 | MSL1, 260 °C reflow (3×) | — | Pass |
Temperature cycling (TC) after precon | −55 to 125 °C | 1000× | Pass |
HAST (w/o bias) after precon | 130 °C/85% RH | 192 h | Pass |
High temperature storage (HTS) | 150 °C | 1000 h | Pass |
TCoB was completed and passed 500 cycles, the typical requirement for consumer and commercial devices, with the results shown in Table 22.3 and the Weibull plot in Figure 22.6. Results were obtained from electrical measurement of daisy‐chain bump structures. The results are comparable with conventional WLCSP products produced with PI dielectrics.
Table 22.3 TCoB reliability test results for eWLCSP.
TCoB (Cond B) | Failure rate | Characteristic life (η) | Weibull slope (β) | First failure |
−40 to 125 °C | 0.635 | 1219.4 | 10.13 | 864× |
Drop test was completed and passed the JEDEC requirement of 30 drops with the results shown in Table 22.4 and the Weibull plot in Figure 22.7. Results were obtained from electrical measurement of daisy‐chain bump structures. The results are comparable with conventional WLCSP products produced with PI dielectrics.
Table 22.4 Drop test results for eWLCSP.
Drop test | Failure rate | Characteristic life (η) | Weibull slope (β) | First failure |
JEDEC | 0.635 | 1553.5 | 5.97 | 772× |
For reliability tests of larger eWLCSP over 6 mm × 6 mm, two test vehicles were prepared, 6 mm × 6 mm and 8 mm× 8 mm, as shown in Table 22.5. The eWLCSP process passed standard reliability tests used in wafer‐level packaging including CLR and BLR (TCoB and drop test). CLR was completed with the test conditions shown in Table 22.6.
Table 22.5 eWLCSP test vehicle (TV) details.
eWLCSP size | Mask no. | Solder ball pitch | |
TV1 | 6 mm × 6 mm | 3 (without UBM) | 0.4 mm |
TV2 | 8 mm × 8 mm | 4 (with UBM) | 0.35 mm |
Table 22.6 Component‐level reliability results.
Component‐level test | Condition | Status | |
MSL1 | MSL1, 260 °C Reflow (3×) | — | Pass |
Temperature cycling (TC) after precon | –55 to 125 °C | 1000× | Pass |
HAST (w/o bias) after precon | 130 °C/85% RH | 192 h | Pass |
High temperature storage (HTS) | 150 °C | 1000 h | Pass |
The evaluation results were confirmed by visual inspection and electrical test. No delamination of the protective coating was detected during the CLR evaluation. TCoB was completed and passed 500 cycles with the results shown in Table 22.7. Results obtained from electrical measurement of daisy‐chain bump structures demonstrate that eWLCSP is comparable with conventional WLCSP products produced with PI dielectrics. Drop test was completed and passed the JEDEC requirement of 30 drops with the results shown in Table 22.4.
Table 22.7 Board‐level reliability test results.
Tests | Conditions | Status |
TCoB | JEDEC JESD22‐A103 –40 to 125 °C |
Pass |
Drop test | JEDEC JESD22‐B111 1500G |
Pass |
A four‐point bending test was carried out to investigate package‐level strength. eWLCSP shows an over 25% increase in die strength compared with WLCSP due to the sidewall protection and optimized backgrinding process.
The Si surface roughness was measured with atomic force microscopy (AFM). eWLCSP has a Si roughness value very close to that of WLCSP. A roughness image scan clearly showed no difference in Si surface roughness between WLCSP and eWLCSP.
The protective sidewall coating is a unique attribute of the eWLCSP. This protective layer is durable and will prevent silicon chipping on the side of the package and has the ability to protect the silicon during socket insertion for test. This has been demonstrated through multiple insertion tests on completed products with no observed damage to the protective coating in Table 22.8.
Table 22.8 Visual inspection results after auto handler socket insertion test.
VM scan yield summary results | Pre VM scan | 1× socket insertion post VM scan | 2× socket insertion post VM scan | 3× socket insertion post VM scan |
# of devices Inspected SOU |
5000 | 5000 | 5000 | 5000 |
# of devices accepted | 5000 | 5000 | 5000 | 5000 |
VM yield (%) | 100.0 | 100.0 | 100.0 | 100.0 |
# of devices rejected | 0 | 0 | 0 | 0 |
With these test results along with CLR and BLR results, large eWLCSP with additional sidewall protection has demonstrated more robust reliability than standard WLCSP and prevents side chip cracking.
An eWLCSP reconstituted wafer is different from a silicon wafer as the backside of the wafer is mold compound, which tends to have a much higher warpage level compared with a silicon wafer of similar thickness. To address this issue, modifications were made to the prober to enable handling of the eWLCSP reconstituted wafer. These modifications were carried out and successfully demonstrated in a high volume manufacturing environment
The test chuck of the prober must be modified for handling of the warped wafers. A progressive and stronger vacuum at the test chuck is necessary to ensure the wafer is properly held flat on the test chunk before testing. The shape of the robot handling arms was modified, and the vacuum to these arms was enhanced to ensure proper holding during the transfer of wafers within each module of the prober. One of the key differences between an eWLCSP reconstituted wafer and a silicon wafer is the wafer identification (ID) marking on the wafer. For an eWLCSP reconstituted wafer, the ID marking is on a copper surface, which makes it look entirely different under the prober’s optical character recognition (OCR) reader. Hence, modification on the OCR reader is required to enable reading of the reconstituted wafer ID. Furthermore, in alignment with the eWLCSP assembly process, where the modified front‐opening unified pod (FOUP) is used in processing the wafers, the prober software must be modified to accommodate the differences. Figure 22.8 shows a pre‐aligner station where the wafer notch is detected and the wafer ID is read of the eWLCSP panel.
Wafer‐level testing of eWLCSP has been proven with noticeable advantages such as a higher test cell utilization and better first pass yield, resulting in an overall cost reduction of testing these packages. Wafer‐level testing offers a short index time, especially for highly parallel testing. This is mainly due to the indexing from one touchdown to the next, where it only involves a small movement within the wafer compared with the pick and place handler that has a high index time. Indeed, this short index time characteristic is most suitable for eWLCSP devices testing, especially for small package sizes, short test times, and high parallelism test requirements. Besides achieving higher throughput, wafer‐level testing improves utilization by lowering manufacturing stoppages, such as jamming associated with the handling of small packages.
Wafer‐level testing of eWLCSP improves the first pass yield by utilizing the visual alignment system, which is standard on a prober. Using camera and vision technology, the socket pin alignment to the package bumps can be highly accurate and repeatable; thus the first pass test yields for wafer‐level testing are significantly higher. Figure 22.9 shows the indexing process of an eWLCSP carrier. Besides providing better and more accurate contacts using vision technology, the prober utilizes its auto socket cleaning feature to improve test yield, whereas this auto cleaning feature is an in‐line process that minimizes test cell downtime and eliminates operator labor needed for manual cleaning. Furthermore, wafer‐level testing also reduces the tooling and hardware costs. In contrast to pick and place handlers that require a new change kit for every different package size, using a prober to handle wafer‐level testing eliminates the need to change kits and thus reduces the overall manufacturing cost.
A new eWLCSP process has been developed and verified with reliability testing. The process provides mechanical sidewall protection to WLCSP parts with an increase in package size of less than 140 μm in X and Y dimensions. The sidewall protection resolves the problem of silicon damage during the assembly process and provides a path to significant cost savings for customers as the panel size is increased. The eWLCSP process described is wafer size agnostic, so the same manufacturing line can process eWLCSP products regardless of the incoming wafer size. 450 mm wafers can easily be accommodated in the eWLCSP process once the service is required by the customers (Figure 22.10).