De-embedding device parasitics of ultra-high speed VCSELs

Abstract

To establish highly performing vertical cavity surface-emitting lasers (VCSELs), it is essential to have an adequate understanding of the intrinsic laser dynamics of these devices. However, this is done while bearing in mind that extrinsic parasitic elements in VCSELs play an important role in limiting the intrinsic modulation bandwidth. In this work, we analyse different electrical parasitic equivalent circuit models in the aim of comparing them and selecting the one that can best describe and represent the physical properties of our high-performance VCSELs. Through measuring the microwave reflection coefficient S11(f), then fitting it with the calculated one from the equivalent circuit impedance model, the parasitic components of the equivalent circuit model can be extracted. The S11(f) data was collected over different ranges of operating bias currents and using a 7 μm oxide aperture diameter VCSEL. This allows us to observe the variations of these circuit elements with respect to the current and compute the transfer function and the resulting parasitic cut-off frequencies (bandwidth limitation) for each model. After plotting and comparing the transfer functions of the different models together, under the same driving current, it was found that the discrepancy between the two curves, in a specific frequency range, is rather small over the VCSEL bandwidth of interest, hence allowing us to use the first-order low pass filter to de-embed the parasitic contributions and separate them from the device intrinsic response. However, over higher frequency ranges, the deviation is found to be substantial and the extract parasitic transfer function should be taken into consideration. Another issue to be addressed is the reliability of the simple circuit models to extract accurate circuit component values, especially when the deviation between the measured microwave reflection coefficient S11(f) and the fitted model is substantially large. This discrepancy is due to the oversimplification imposed on the equivalent circuit model, leading to a high level of uncertainty in the extracted circuit component values. Thus, sufficient modelling and accurate fitting strategies are needed for a reliable parasitic de-embedding approach.