Solutions to Optoelectronic packaging problems

Optoelectronic bonding has been performed onto sample MCMs (Multi-Chip Modules), which are to be used for optical testing for the High-Speed Optoelectronic Memory Systems [HOLMS] demonstrator. This was conducted on an FC6 bonder using the flip chip bonding process. Initially, the photodiodes [PDs] and Laser diodes [LDs] were successfully bonded, but we experienced problems with attaching the electronic Mixed Signal chips [MSC]. Photodetectors [PDs]: For the HOLMS project, 1×12 arrays of PIN diodes with 180 µm optical windows at 250 µm pitch running at 1.2 GHz will be used. These will match the VCSEL array. Its operation will be at 850 nm with an expected sensitivity of 0.5/0.6 AW^-1. The photodiodes were constructed on a GaAs/AlGaAs strained layer structure. Also, the PDs have two layouts of 1×12 arrays, i.e. shoe and ring designs. HWU engaged with EPSRC National Centre for III-V Technologies located at University of Sheffield to produce a suitable detector wafer. VCSELs: 1×12 arrays at 250 µm pitch capable of a minimum 2.5 GHz with a wavelength of 850nm were selected. The devices have a threshold current of 1.8 mA and work with an operating current of 5 mA. Initially, devices with a typical optical output of 1mW at 3mA and thickness 150 µm were obtained, but later VCSELs, which had a higher optical power of 2 mW at 3 mA and thickness 200 µm with the same specifications as the earlier version, were also acquired. Test chips: The Mixed Signal Chip was designed using the 0.35 µm BiCMOS SiGe technology with a power supply voltage of 3.3 Volts. The function of this chip was to provide a test bed to assess the performances of the whole link: electrical - optical - electrical. The chip will provide self test features as provision for error count and can be configured as an emitter or receiver. An experimental set-up was constructed in order to characterise the PDs. This was performed by using a probe station connected to the HP 4101 parameter analyser. Tests to find dark current values and optical testing using an 850 nm source was successfully conducted. Dark measurements obtained for both shoe and ring detector arrays were ~ 10^-10 Amps, and calculated series resistance for diodes ranged between 26-33 O. These values were consistent with the parameters given by the manufacturer. Generated photocurrent data obtained from optical testing measurements was used to deduce the Responsivity [R] of the detectors. Hence, calculated R ranged between 0.37-0.42AW ^-1. Overall, the shoe devices showed more consistency in its results in comparison with the ring layout. Hence, a decision was made to choose the shoe design for implementation into the HOLMS demonstrator. The optoelectronic devices were successfully assembled onto the test modules by means of conductive bumps, i.e. MSCs & PDs using gold (Au) stud and VCSELs with AuSn. For the photodiodes and chips, the bumps were placed on the MCM and for the Laser Diodes - they were bumped by the manufacturer. A number of fully populated MCMs consisting of optoelectronic devices were produced and then soldering of passive components was conducted, i.e. resistors, capacitors, inductors and connectors. The next step is to perform optical testing on the completed MCMs. Originally, in our first trials, test chips were bonded using the ACF process (Anisotropic Conductive Film). Unfortunately, this method was discontinued as we achieved no electrical connectivity between chip and substrate and also the FC6 bonder can't apply the required force to bond ~700 pads on the final HOLMS chip. Therefore, Au stud bumping was used as the flip chip bonding technology. The bumps were placed on respective footprints located on test modules, and then the test chips were bonded using IR thermocompression process. After flip chip bonding the electronic devices, the following observation was seen: (1) the devices initially bonded onto the substrate, but after thermal cooling, they became detached from the module or (2) they did not bond. We concluded that the reason why this occurred was due to the rapid oxidation of the AlCu pads on the Mixed Signal Chip. The process is that Al oxidises very quickly, and forms an oxidation layer on the pad surface. This extra layer inhibits a flip chip connection with the substrate. Hence, the result is that test chips will not bond onto the MCM. Therefore, plasma cleaning of the test chips was conducted. Devices were then flip chip bonded onto test modules using the scrubbing function. The results were positive, and successful bonding was achieved. Hence, the other components, i.e. PDs and VCSELs were attached onto the substrates. Initially, we were intending to use AuSn solder bump technology for flip chip bonding our components. Hence, bonding the photodiodes and VCSELs using AuSn bumps is feasible, but there was a major issue with attaching the chips. The metallisation of the pads on chips are TiN - Al alloy. Hence, this surface is not suitable for flip chip connection using AuSn bumps. It is recommended that to achieve AuSn solder ability, the top layer of the pad has to be Au. Therefore, Ni/Au or Ni/Pd/Au metallisation on the bonding pads of chips would be eligible. Also, another problem that would result is the diffusion of Sn with Cu forming SnCu intermetallics. This would cause fatigue and weaken the solder joint, therefore resulting in a loss of connection. The metallization of bonding pads for the final HOLMS chips can't be changed as the manufacturer's fabrication process is unable to be modified. We are looking into using another solder bump material, i.e. SnPb or lead free Sn95,5Ag4CuO,5. If it is concluded that solder technology is not suitable for attaching the final chips we do have the option of using Au stud bump flip chip technology as we showed this was successful. But care, must be taken to minimise or possibly eliminate the oxidation of Aluminium pads. The fully populated MCMs will undergo optical testing. Tests will be conducted on: (1) optical clock signals, (2) measurements of receiver sensitivity using clock signals and (3) bit error rate as a function of the input power on the photodiode. The experiments will be performed using two separate MCMs, where VCSEL output from one MCM will be directed to a photodetector of another module. © 2005 IEEE.