Page 51 - Chip Scale Review_January February_2022-digital
P. 51

FOWLP and Si-interposer for high-speed


        photonics packaging


        By Lim Teck Guan, Eva Wai Leong Ching, Jong Ming Ching, Loh Woon Leng, David Ho Soon Wee, Sajay B G, et al.
        [Institute of Microelectronics, A*STAR (Agency for Science, Technology and Research)]
        T        he ever-increasing demand   Dat a center OEs use ICs f rom     IME is currently developing two




                 for higher data bandwidth
                                           SiGe for drivers and amplifiers, and
                                                                              out wafer-level packaging (FOWLP)
                 in the data center has led   diverse technologies such as CMOS/  packaging platforms based on fan-
        to the requirement of higher speed,   III-V lasers. These ICs need to be   and the Si-interposer to address the
        smaller form factors and scalable,   integrated into very small form factor   EIC and PIC integration requirement.
        integrated optical engines (OE). An OE   system-in-packages (SiPs) to be able   The first solution is based on the
        comprises various electronic integrated   to bring in optical signals and convert   established FOWLP technology, which
        circuits (EICs) and photonics ICs   them into electrical signals (and vice   has been demonstrated for digital
        (PICs) to enable the optical/electrical   versa) to be processed by switches,   and radio-frequency  (RF)/mmWave
        (O/E) signal conversion. With the   FPGAs or other application-specific   applications [3] integration. The
        advancement in semiconductor device   ICs (ASICs). Therefore, electronic-  FOWLP has good RF performance
        technology, packaging and integration   photonic heterogeneous integration   and allows multi-chip heterogeneous
        technologies are becoming the limiting   enabled by advanced wafer-level   integration to be well suited for this
        factors to enable the OE to meet the   packaging is a promising technique to   high-performance OE  application.
        demand of the data center’s high data   realize high-speed optical engines for   For  the  Si-interposer  with  a  high-
        bandwidth requirement of more than   the data center.                 resistivit y subst rate, it can also
        800Gbps. It is, therefore, important   For the best per for mance with   provide very high-speed and high-
        to develop a cost-effective PIC and   respect to speed, the EIC is integrated   bandwidth PIC and EIC integration.
        EIC packaging platform to realize a   directly on top of the PIC. Currently,   Additionally, it can provide a sub-
        high data rate OE. The integration and   silicon photonics is the most promising   micron alignment feature for the fiber
        packaging technologies must support   tech nolog y t hat ca n provide t he   to the PIC passive assembly.
        the scaling of the number of optical   necessary performance and highest
        channels and provides a high-speed   functionality integration. The PIC here   Photonics FOWLP
        electrical interconnect between the   is an active interposer—it not only   The FOWLP integration platform
        PIC and EIC of each channel of more   consists of the photonics circuit, but also   w e d e s c r i b e i s  f o r h i g h - s p e e d
        than 100Gbps.                      provides the necessary physical area   PIC and EIC integration. It is low
          Advanced wafer-level packaging   for the EIC and external routing and   c o s t  a nd  it  le ve r a ge s a d v a n c e d
        has been successfully used in state-  integration [1]. However, this is a costly   FOWLP development for electronic
        of-the-art field-programmable gate   solution because the PIC will require   semiconductor packaging. Currently,
        a r r a y ( F P GA) IC s , s m a r t pho n e   an additional large area to support   only the PIC with vertical coupling
        application processors, and graphics   the EIC. In addition, for high-speed   in FOWLP has been demonstrated
        processing units (GPUs) to provide   integration, through-silicon vias (TSVs)   [4]. The integration of the edge PIC
        power-performance-form factor boosts   will be required to be used for the PIC   remains a big challenge because
        that are not obtained by conventional   active interposer. The fabrication of the   the embedding mold compound is
        packaging. A key benefit of advanced   TSV on the PIC requires many complex   not transparent in the optical signal
        w a f e r - l e v e l  p a c k a g i n g i s t h e   process steps as described in [2].   spect r um, and the optical sig nal
        capability to achieve heterogeneous   Furthermore, the Si photonics circuit is   cannot be coupled from the optical I/
        integration whereby ICs from diverse   fabricated in a 200mm wafer instead of   O at the edge of the PIC after molding.
        technologies (complementary metal-  a 300mm wafer, which makes the cost   The  optical  I/O  is  very  sensitive  to
        o x i de   s e m i c o n d uc t o r   [C M O S ] ,   of this integration solution unattractive   contamination  because the optical
        SiGe, silicon-on-insulator [SOI],   for most  commercial applications. The   signal has a very short wavelength
        PIC,  III-V  semiconductors,  etc.)   alternative solution is using various   (~1. 3μ m   t o   ~1. 5μ m),   t h e r efo r e ,
        ca n be i ndependently opt i m i zed   complex customized multiple assembly   polishing and cleaning the package
        and tightly integrated in small form   technologies for the multi-chip module   to reveal the PIC edge I/O is very
        factor packages to achieve power-  integration. However, because of the   challenging and time consuming. To
        p e r f or m a n c e - f or m  fa c t or- c o s t   complex assembly requirement and the   overcome this problem, the PIC optical
        optimization that is not otherwise   limited scaling capability, this solution   I/O at the edge will be designed with
        possible with monolithic integration.  is not preferred.              an additional silicon buffer section, as


                                                                                                             49
                                                          Chip Scale Review   January  •  February  •  2022   [ChipScaleReview.com]  49
   46   47   48   49   50   51   52   53   54   55   56