Xin Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Xin Wang has a total of 47 co-authors. Here are the top co-authors by number of shared publications:
- Achim Vandierendonck
Ghent University
3 shared publications - Guy Torfs
Ghent University
3 shared publications - Johan Bauwelinck
Ghent University - IMEC
3 shared publications - Caro Meysmans
Ghent University
2 shared publications He Li
Ghent University - IMEC
2 shared publications- Senbiao Qin
Ghent University - IMEC
2 shared publications
Latest Publications
- 29 May 2026
- Published by Optica Publishing Group in Optics Express
- Vol. 34 (11) , 20833-20842
56-Gb/s 3D-integrated silicon optical transmitter based on micro-transfer printing
We present an optical transmitter consisting of a SiGe BiCMOS driver amplifier heterogeneously integrated through micro-transfer printing onto a silicon photonic chip that contains a Mach-Zehnder modulator (MZM) and optical couplers. The driver has a footprint of only 200 µm × 300 µm, and features small bond pads of 20 µm × 20 µm. It was fabricated using a modified SiGe BiCMOS on SOI process to enable the release process prior to micro-transfer printing. The post-print metallization process to contact the EIC with the PIC consists of spin coating a layer of divinylsiloxane-bis-benzocyclobutene (DVS-BCB) to create a ramp to overcome the 20 µm step height, after which a gold metal layer is deposited to connect the driver to the MZM. To verify the operation of the optical transmitter, a high-speed measurement setup is used to capture eye diagrams. The 3D-integrated transmitter achieves 56 Gb∕s non-return to zero (NRZ) with a 5-tap feed-forward equalizer (FFE) and 28 GBd 4-level pulse amplitude modulation (PAM-4) with a 5-tap FFE and nonlinearity pre-compensation, both at a total transmitter power efficiency of 9.1 pJ∕bit. Integration through micro-transfer printing offers the possibility of greatly increasing scalability and fabrication throughput of short-reach optical transceivers for data centers.
- 29 May 2026
- Published by Optica Publishing Group in Optics Express
- Vol. 34 (11) , 20833-20842
56-Gb/s 3D-integrated silicon optical transmitter based on micro-transfer printing
We present an optical transmitter consisting of a SiGe BiCMOS driver amplifier heterogeneously integrated through micro-transfer printing onto a silicon photonic chip that contains a Mach-Zehnder modulator (MZM) and optical couplers. The driver has a footprint of only 200 µm × 300 µm, and features small bond pads of 20 µm × 20 µm. It was fabricated using a modified SiGe BiCMOS on SOI process to enable the release process prior to micro-transfer printing. The post-print metallization process to contact the EIC with the PIC consists of spin coating a layer of divinylsiloxane-bis-benzocyclobutene (DVS-BCB) to create a ramp to overcome the 20 µm step height, after which a gold metal layer is deposited to connect the driver to the MZM. To verify the operation of the optical transmitter, a high-speed measurement setup is used to capture eye diagrams. The 3D-integrated transmitter achieves 56 Gb∕s non-return to zero (NRZ) with a 5-tap feed-forward equalizer (FFE) and 28 GBd 4-level pulse amplitude modulation (PAM-4) with a 5-tap FFE and nonlinearity pre-compensation, both at a total transmitter power efficiency of 9.1 pJ∕bit. Integration through micro-transfer printing offers the possibility of greatly increasing scalability and fabrication throughput of short-reach optical transceivers for data centers.
- 16 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Journal of Lightwave Technology
- Vol. PP (07338724) , 1-14
Sub-pJ/b 28 nm CMOS Inverter-Based Direct-Drive Optical Tx Using 1060 nm SM VCSEL Enabling 60 Gbaud NRZ Transmission Over 5 km SMF
This article presents a direct-drive optical transmitter (Tx) using common-cathode (CC) type single-mode (SM) 1060 nm vertical-cavity surface-emitting laser (VCSEL) suitable for medium-reach transmission (< 5 km) for linear pluggable optics (LPO)-based inter-cluster datacenter interconnect. The VCSELdriver featuring 4 channels is implemented in 28 nm bulk complementary metal-oxide semiconductor (CMOS) technology and co-designed with a GaAs-based 1060 nm VCSEL array. A VerilogA-based VCSEL device model with self-heating-induced voltage and optical power adjustments is introduced to facilitate co-design and (co-)optimization electro-optical interface. Two CMOS inverter-based equalizers, based on an active inductor and a T-coil, are used to enhance the driver bandwidth. A group-delay compensator is added to compensate the VCSEL group-delay variation and improve in-band gain flatness. The VCSEL features an optimal-3 dB BW of 31 GHz with 6 mA current bias. The optical Tx successfully demonstrated 60 Gbaud non-return-to-zero (NRZ) transmission with a SM 1060 nm VCSEL over 5 km single-mode fiber (SMF) using a 10-tap feed-forward-equalizer (FFE) with a record-high FEC-corrected bitrate-distance product of 283.47 Gb/s.km. For the electro optical transmission experiment, the optical Tx achieves a Tx energy-efficiency of 0.94 pJ/b with a 6 mA current biased VCSEL having-2.7 V supply at the CC terminal.
- 16 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Journal of Lightwave Technology
- Vol. PP (07338724) , 1-14
Sub-pJ/b 28 nm CMOS Inverter-Based Direct-Drive Optical Tx Using 1060 nm SM VCSEL Enabling 60 Gbaud NRZ Transmission Over 5 km SMF
This article presents a direct-drive optical transmitter (Tx) using common-cathode (CC) type single-mode (SM) 1060 nm vertical-cavity surface-emitting laser (VCSEL) suitable for medium-reach transmission (< 5 km) for linear pluggable optics (LPO)-based inter-cluster datacenter interconnect. The VCSELdriver featuring 4 channels is implemented in 28 nm bulk complementary metal-oxide semiconductor (CMOS) technology and co-designed with a GaAs-based 1060 nm VCSEL array. A VerilogA-based VCSEL device model with self-heating-induced voltage and optical power adjustments is introduced to facilitate co-design and (co-)optimization electro-optical interface. Two CMOS inverter-based equalizers, based on an active inductor and a T-coil, are used to enhance the driver bandwidth. A group-delay compensator is added to compensate the VCSEL group-delay variation and improve in-band gain flatness. The VCSEL features an optimal-3 dB BW of 31 GHz with 6 mA current bias. The optical Tx successfully demonstrated 60 Gbaud non-return-to-zero (NRZ) transmission with a SM 1060 nm VCSEL over 5 km single-mode fiber (SMF) using a 10-tap feed-forward-equalizer (FFE) with a record-high FEC-corrected bitrate-distance product of 283.47 Gb/s.km. For the electro optical transmission experiment, the optical Tx achieves a Tx energy-efficiency of 0.94 pJ/b with a 6 mA current biased VCSEL having-2.7 V supply at the CC terminal.
- 6 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Journal of Lightwave Technology
- Vol. 44 (12) , 5048-5058
APD Receiver Co-Optimization for Chip-to-Chip Communication on 3D/2.5D Photonic Interposers
This work presents the analysis and optimization of a Ge/Si avalanche photodiode (APD) receiver implemented in a 22 nm silicon-on-insulator (SOI) technology node for 3D CMOS photonic links. The fabricated APD achieves a gain–bandwidth product of 250 GHz under a reverse bias of 9 V. Post-layout simulations results show that the receiver improves the optical modulation sensitivity by up to 3 dB compared to a PIN receiver with post-amplification. The APD must be operated in the reverse bias range of $[3,9]\,\text{V}$ to keep the intersymbol interference (ISI) power penalty at 0 dB. These results are obtained through a co-optimization analysis that evaluates the trade-offs between bandwidth, noise, gain, and power consumption of the APD receiver blocks, including the analog front end and sampling stages.
- 6 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Journal of Lightwave Technology
- Vol. 44 (12) , 5048-5058
APD Receiver Co-Optimization for Chip-to-Chip Communication on 3D/2.5D Photonic Interposers
This work presents the analysis and optimization of a Ge/Si avalanche photodiode (APD) receiver implemented in a 22 nm silicon-on-insulator (SOI) technology node for 3D CMOS photonic links. The fabricated APD achieves a gain–bandwidth product of 250 GHz under a reverse bias of 9 V. Post-layout simulations results show that the receiver improves the optical modulation sensitivity by up to 3 dB compared to a PIN receiver with post-amplification. The APD must be operated in the reverse bias range of $[3,9]\,\text{V}$ to keep the intersymbol interference (ISI) power penalty at 0 dB. These results are obtained through a co-optimization analysis that evaluates the trade-offs between bandwidth, noise, gain, and power consumption of the APD receiver blocks, including the analog front end and sampling stages.
- 10 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 61 (1) , 318-330
A Duty-Cycle Switching 30-Gb/s Burst-Mode CDR With 1.6-ns Locking Time in 28-nm CMOS
This article presents a closed-loop type burst-mode clock and data recovery (BM-CDR) circuit with fast phase offset detection using 8/3x-fractional oversampling in the periodic preamble. The proposed phase offset detector achieves a resolution of 1/8 unit intervals (UIs) with a detection time of 4 UIs. A 2x-oversampling closed-loop bang-bang CDR is performed after the phase offset detection to provide jitter filtering. The switching between these two different oversampling ratios is realized in a single multi-phase clock generator (MPCG) by changing the duty-cycle in four differential quarter-rate clocks. Furthermore, fast duty-cycle switching (DCS) is introduced in the injection-locked ring oscillator (ILRO) design to speed up the transition from one sampling ratio to another. A prototype fabricated in 28-nm CMOS achieves a locking time of 1.6 ns at 30-Gb/s data rate, a BER of 1E-12 with a recovered clock integrated rms jitter of 398.4 fs. The jitter tolerance curve shows a corner frequency around 20 MHz with a 20-dB/dec slope in the low-frequency region. The receiver including the proposed CDR consumes 75.53 mW with 0.9-V supply and occupies an area of 0.148 mm2.
- 10 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 61 (1) , 318-330
A Duty-Cycle Switching 30-Gb/s Burst-Mode CDR With 1.6-ns Locking Time in 28-nm CMOS
This article presents a closed-loop type burst-mode clock and data recovery (BM-CDR) circuit with fast phase offset detection using 8/3x-fractional oversampling in the periodic preamble. The proposed phase offset detector achieves a resolution of 1/8 unit intervals (UIs) with a detection time of 4 UIs. A 2x-oversampling closed-loop bang-bang CDR is performed after the phase offset detection to provide jitter filtering. The switching between these two different oversampling ratios is realized in a single multi-phase clock generator (MPCG) by changing the duty-cycle in four differential quarter-rate clocks. Furthermore, fast duty-cycle switching (DCS) is introduced in the injection-locked ring oscillator (ILRO) design to speed up the transition from one sampling ratio to another. A prototype fabricated in 28-nm CMOS achieves a locking time of 1.6 ns at 30-Gb/s data rate, a BER of 1E-12 with a recovered clock integrated rms jitter of 398.4 fs. The jitter tolerance curve shows a corner frequency around 20 MHz with a 20-dB/dec slope in the low-frequency region. The receiver including the proposed CDR consumes 75.53 mW with 0.9-V supply and occupies an area of 0.148 mm2.
- 9 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Journal of Lightwave Technology
- Vol. 43 (4) , 1595-1603
A Bi-Directional Distributed Multi-User MIMO Testbed Using Digital Sigma-Delta-Over-Fiber
Sigma-Delta modulated signals over fiber is proposed for fronthauling of distributed signals. (AROF) provides perfect synchronization, but linearity is directly impacted by the required circuit elements. (DRoF) has a low cost of implementation but complicates synchronization. Sigma-Delta modulated signals over fiber combines the synchronization benefit from ARoF and the low cost of DRoF. In this work, we show a simplified and synchronous distributed multiple-input multiple-output (MIMO) communication system for both up-link and down-link transmission, based on our real-time sigma-delta modulated signals over fiber front-haul network. The testbed uses a carrier frequency of 3.686 GHz and is composed of 4 distributed units with 4 antennas each, totaling 16 antennas. The received clock is used to provide synchronization between the central and distributed units. With this implementation, we demonstrate distributed multi-user MIMO. The synchronicity of the system is demonstrated by performing over-the-air channel reciprocity calibration and reciprocity-based uplink transmission. This is verified using 64-QAM OFDM symbols. Measurement results for 2 users / 4 users show an average error vector magnitude for up-link of 2.51% (-32.02 dB) / 2.89% (-30.79 dB) and for down-link of 4.23% (-27.48 dB) / 6.09% (-24.31 dB) respectively.
- 9 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Journal of Lightwave Technology
- Vol. 43 (4) , 1595-1603
A Bi-Directional Distributed Multi-User MIMO Testbed Using Digital Sigma-Delta-Over-Fiber
Sigma-Delta modulated signals over fiber is proposed for fronthauling of distributed signals. (AROF) provides perfect synchronization, but linearity is directly impacted by the required circuit elements. (DRoF) has a low cost of implementation but complicates synchronization. Sigma-Delta modulated signals over fiber combines the synchronization benefit from ARoF and the low cost of DRoF. In this work, we show a simplified and synchronous distributed multiple-input multiple-output (MIMO) communication system for both up-link and down-link transmission, based on our real-time sigma-delta modulated signals over fiber front-haul network. The testbed uses a carrier frequency of 3.686 GHz and is composed of 4 distributed units with 4 antennas each, totaling 16 antennas. The received clock is used to provide synchronization between the central and distributed units. With this implementation, we demonstrate distributed multi-user MIMO. The synchronicity of the system is demonstrated by performing over-the-air channel reciprocity calibration and reciprocity-based uplink transmission. This is verified using 64-QAM OFDM symbols. Measurement results for 2 users / 4 users show an average error vector magnitude for up-link of 2.51% (-32.02 dB) / 2.89% (-30.79 dB) and for down-link of 4.23% (-27.48 dB) / 6.09% (-24.31 dB) respectively.
- 18 January 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Micro
- Vol. 43 (2) , 86-95
Photonic Network-on-Wafer for Multichiplet GPUs
This paper introduces the Photonic Network-on-Wafer (NoW) GPU architecture to overcome fundamental limitations in electrical interconnect scaling by implementing the inter-GPU network in a wafer-scale optical interposer. We argue that the photonic-NoW GPU is a scalable architecture, delivering significant performance benefits in a power-efficient manner.
- 18 January 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Micro
- Vol. 43 (2) , 86-95
Photonic Network-on-Wafer for Multichiplet GPUs
This paper introduces the Photonic Network-on-Wafer (NoW) GPU architecture to overcome fundamental limitations in electrical interconnect scaling by implementing the inter-GPU network in a wafer-scale optical interposer. We argue that the photonic-NoW GPU is a scalable architecture, delivering significant performance benefits in a power-efficient manner.