Contents I. Introduction
The exponential growth of data traffic with the thriving of cloud services and wireless networks calls for high-speed data transmission over short distances in local area networks (LANs) and data centers. To cope with the huge data traffic, the data rate of Ethernet is migrating from 100 Gb/s (100 GbE) toward 400 Gb/s (400 GbE) [1]. To ensure the 56 Gb/s modulation required for 400 GbE based on the 4 × 100 Gb/s scheme, light sources should have a modulation bandwidth exceeding 40 GHz to ensure desirable performances. Compared with directly modulated lasers (DMLs) [2], [3], electroabsorption modulated lasers (EMLs), i.e., distributed feedback (DFB) lasers monolithically integrated with electroabsorption (EA) modulators, are more advantageous for 400 GbE systems, as they exhibit higher extinction ratio, more flattened frequency response and lower frequency chirp [4], [5]. The bandwidth of an EML is mainly limited by the capacitance of the EA modulator. Though travelling-wave EA modulators have been proposed to overcome the RC limitation [6], [7], and bandwidth over 100 GHz has been demonstrated [8], the complicated electrode structure has limited their wide applications. On the other hand, with careful device design and fabrication, lumped-electrode EMLs for 400 GbE have been demonstrated [9]–[11]. Apart from the device capacitance, parasitics due to wire bonding also have a crucial influence on the frequency response of a lumped-electrode EML. To remedy the bandwidth degradation due to bonding-wire induced parasitic inductance, flip-chip interconnection has been adopted in place of bonding-wires to connect the RF circuit board and the EML chip [12], [13]. And lumped-electrode EMLs based on flip-chip interconnection can achieve a bandwidth of 59 GHz and 107-Gb/s modulation rate under 4-intensity-level pulse amplitude modulation (PAM4) [14].
