Photodetectors (PDs) are an important active device in optoelectronic integrated circuits (OEICs), and, for shorter haul interconnections where circuit (e.g. transimpedance amplifier (TIA)) noise may be the dominant noise in receivers, metal-semiconductor-metal photodiodes (MSM PDs) are attractive due to their low capacitance per unit area compared to PIN photodetectors and the ease of monolithic integration with field effect transistors (FETs). Inverted-MSM PDs (I-MSM PDs), which are thin film MSM PDs with the fingers on the bottom of the device, have demonstrated higher responsivities compared to conventional MSM PDs while maintaining small capacitance per unit area, low dark current (~nA), and high speed. However, the modeling of MSM PDs and I-MSM PDs for insertion into circuit simulators for integrated PD/TIA modeling has not been reported. In this paper, an accurate high-frequency equivalent circuit-level model of thin film I-MSM PDs is obtained using an on-wafer measurement-based modeling technique. This circuit-level model of MSM PDs can be used for capacitance sensitive preamplifier design for co-optimization with widely used simulators (ADS and HSPICE). The obtained circuit-level model shows good agreement with measured s-parameters.
Linear statistical models have been generated to predict the performance of metal-semiconductor-metal (MSM) PDs for multi-gigabit optical interconnections. The models estimate the bandwidth and responsivity of the MSM PDs based on the input factors: absorbing layer thickness, detector size, finger widths and finger gaps. The design of experiments (DOE) approach was employed to obtain the necessary data to construct the models.
Numerous samples were fabricated so that multiple devices measurements could serve to both construct and verify the linear statistical models. The MSM PDs were fabricated from material with structure InAlAs/InAlGaAs/InGaAs (2000Å, 3000Å or 5000Å, absorbing layer)/InAlAs. The MSM interdigitated fingers were photolithographically defined with finger gaps and widths varying as DOE parameters. A benzocyclobutene (BCB, Cyclotene 35) layer was spin-coated onto all of the samples as isolation from the probing pads.
In the bandwidth analysis, the detector size (S) and material thickness (T) were investigated with a fixed finger width (1 μm) and gap (1 μm). Taking the measured results of these detectors in the design matrix, and using least square regression, the model equations were derived as: Bandwidth (GHz) = 12.87 - 0.065S - 3T - 0.02ST. After these equations were developed, predictive calculated results from these equations were then further used to predict and compare measured results on devices that were not used in the statistical model. This leads to an average deviation between predicted and measured bandwidth of less than 5%. In the responsivity analysis, the predictive calculation leads to an average deviation less than 11%.