微电子技术使电子器件从分立走向集成，带来成本、可靠性、功耗和体积等方面长达几十年的指数级改善，是电子信息技术能够得到广泛应用的关键。当前，电子信息技术正遭遇带宽和能耗等瓶颈制约，摩尔定律步履维艰。光子作为另一种主要的信息载子，具有高带宽、高速率、低功耗和高并行等特性，信息技术的继续进步必须更为倚重光子。但是，光子器件的成本、可靠性和规模生产性等方面严重落后于电子器件。因此，基于电子器件发展的历史经验，我们提出光电子“微电子化”，强调光电子以“集成”为发展轨道，以“光电融合”为发展方向。以集成为发展轨道，是强调光电子需借鉴微电子的经验，以适应现代信息系统对器件的要求。光电子的集成化发展有3个主要特征：一是芯片平台硅基化，二是集成规模稳步提升，三是生产模式向fabless演进。以光电融合为发展方向，是强调光电子需与微电子相互结合，通过两者的器件一体化、功能融合化来共同解决信息技术目前面临的带宽、速率和功耗等挑战，以适应数字孪生和元宇宙等应用的要求。光电融合主要有芯片和系统两个层面，涉及到单片光电集成、光电共封装（CPO）、混合光电集成、设备解汇聚和光电混合计算等。光电子“微电子化”不仅是一种技术发展指引，还将对信息光电子产业形成重要影响。文章从市场、竞争、企业和产品等角度进行了开放式思考，指出其可能引发产业重塑。信息光电子发展的“微电子化”是信息技术继续前进的客观要求。一方面，光子正由传输技术泛化为信息通信技术（ICT）全域的泛在连接技术，宏尺度上从地面进入海洋和太空，微尺度上从架间、板间进入板内、封装内和片内。另一方面，光子将由带宽提供技术泛化为ICT全域的硬件基础技术，从传输和连接进入计算、处理和路由等复杂功能域。光电子的“微电子化”将使光子不仅有潜力而且有能力充分发挥自身优势以支撑信息技术的继续进步。

Integrated photonics provides a promising platform for quantum key distribution (QKD) system in terms of miniaturization, robustness, and scalability. Tremendous QKD works based on integrated photonics have been reported. Nonetheless, most current chip-based QKD implementations require additional off-chip hardware to demodulate quantum states or perform auxiliary tasks such as time synchronization and polarization basis tracking. Here, we report a demonstration of resource-efficient chip-based BB84 QKD with a silicon-based encoder and a decoder. In our scheme, the time synchronization and polarization compensation are implemented relying on the preparation and measurement of the quantum states generated by on-chip devices; thus, we need no additional hardware. The experimental tests show that our scheme is highly stable with a low intrinsic quantum bit error rate of 0.50%±0.02% in a 6 h continuous run. Furthermore, over a commercial fiber channel up to 150 km, the system enables the realization of secure key distribution at a rate of 866 bit/s. Our demonstration paves the way for a low-cost, wafer-scale manufactured QKD system.

Based on the 90 nm silicon photonics commercial foundry, sidewall-doped germanium–silicon photodetectors (PDs) are designed and fabricated. The large designed overlap between the optical field and electric field achieves high responsivity while retaining high-speed performance. Even including the loss due to optical fiber coupling, the PD demonstrates an external responsivity greater than 0.55 A/W for transverse magnetic (TM) polarization and 0.65 A/W for transverse electric (TE) polarization at 1530 nm. A flat responsivity spectrum of >0.5A/W is achieved up to 1580 nm for both polarizations. Their internal responsivities can exceed 1 A/W in the C+L optical communication bands. Furthermore, with the aid of a 200 mm wafer-level test and analysis, the overall PDs of 26 reticles have a 3 dB optoelectrical bandwidth >50GHz and a dark current <10nA at a -3V bias voltage. Finally, the eye diagram performances under TE and TM polarizations, various modulation formats, and different input wavelengths are comprehensively investigated. The clear open electrical eye diagrams up to 120, 130, 140, and 150 Gbit/s nonreturn-to-zero are experimentally attained at a photocurrent of 1 mA. To the best of our knowledge, this is the first time that single-lane direct detection of record-high-speed 200, 224, 256, and 290 Gbit/s four-level pulse amplitude modulation (PAM) and 300, 336, 384, and 408 Gbit/s eight-level PAM optical signals has been experimentally achieved.

人工神经网络（ANN）学科的发展推动了信息处理技术的进步，目前，被称为第3代ANN的脉冲神经网络（SNN）以其更具生物可解释性，更适合ANN硬件实现的优势受到业界的广泛关注，并已成功应用于模式识别、医学成像和智能控制等多个领域。受制于“后摩尔时代”电子芯片的制程不断接近极限以及冯·诺依曼体系“存算分离”带来的性能瓶颈，低时延、低能耗、高带宽和高并行性的光子计算方案应用于SNN的硬件实现成为信息处理领域多学科融合的热门课题。文章介绍了光子SNN的起源，利用光学器件的特性实现神经元的行为和突触连接强度的变化进而实现SNN的研究历程和多种实现方案，总结了光子SNN目前存在的瓶颈和挑战，展望了光子SNN的未来发展趋势。

Parity‐time (PT) symmetry breaking offers mode selection capability for facilitating single‐mode oscillation in the optoelectronic oscillator (OEO) loop. However, most OEO implementations depend on discrete devices, which impedes proliferation due to size, weight, power consumption, and cost. In this work, we propose and experimentally demonstrate an on-chip tunable PT‐symmetric OEO. A tunable microwave photonic filter, a PT‐symmetric mode‐selective architecture, and two photodetectors are integrated on a silicon‐on‐insulator chip. By exploiting an on‐chip Mach–Zehnder interferometer to match the gain and loss of two mutually coupled optoelectronic loops, single‐mode oscillation can be obtained. In the experiment, the oscillation frequency of the on-chip tunable PT‐symmetric OEO can be tuned from 0 to 20 GHz. To emulate the integrated case, the OEO loop length is minimized, and no extra-long fiber is used in the experiment. When the oscillation frequency is 13.67 GHz, the single‐sideband phase noise at 10-kHz offset frequency is -80.96 dBc / Hz and the side mode suppression ratio is 46 dB. The proposed on-chip tunable PT‐symmetric OEO significantly reduces the footprint of the system and enhances mode selection.

A light-trapping-structure vertical Ge photodetector (PD) is demonstrated. In the scheme, a 3 μm radius Ge mesa is fabricated to constrain the optical signal in the circular absorption area. Benefiting from the light-trapping structure, the trade-off between bandwidth and responsivity can be relaxed, and high opto-electrical bandwidth and high responsivity are achieved simultaneously. The measured 3 dB bandwidth of the proposed PD is around 67 GHz, and the responsivity is around 1.05 A/W at wavelengths between 1520 and 1560 nm. At 1580 nm, the responsivity is still over 0.78 A/W. A low dark current of 6.4 nA is also achieved at -2V bias voltage. Based on this PD, a clear eye diagram of 100 GBaud four-level pulse amplitude modulation (PAM-4) is obtained. With the aid of digital signal processing, 240 Gb/s PAM-4 signal back-to-back transmission is achieved with a bit error ratio of 1.6×10-2. After 1 km and 2 km fiber transmission, the highest bit rates are 230 and 220 Gb/s, respectively.

In recent years, optical modulators, photodetectors, (de)multiplexers, and heterogeneously integrated lasers based on silicon optical platforms have been verified. The performance of some devices even surpasses the traditional III-V and photonic integrated circuit (PIC) platforms, laying the foundation for large-scale photonic integration. Silicon photonic technology can overcome the limitations of traditional transceiver technology in high-speed transmission networks to support faster interconnection between data centers. In this article, we will review recent progress for silicon PICs. The first part gives an overview of recent achievements in silicon PICs. The second part introduces the silicon photonic building blocks, including low-loss waveguides, passive devices, modulators, photodetectors, heterogeneously integrated lasers, and so on. In the third part, the recent progress on high-capacity silicon photonic transceivers is discussed. In the fourth part, we give a review of high-capacity silicon photonic networks on chip.

A quantum dot (QD) mode-locked laser as an active comb generator takes advantage of its small footprint, low power consumption, large optical bandwidth, and high-temperature stability, which is an ideal multi-wavelength source for applications such as datacom, optical interconnects, and LIDAR. In this work, we report a fourth-order colliding pulse mode-locked laser (CPML) based on InAs/GaAs QD gain structure, which can generate ultra-stable optical frequency combs in the O-band with 100 GHz spacing at operation temperature up to 100°C. A record-high flat-top optical comb is achieved with 3 dB optical bandwidth of 11.5 nm (20 comb lines) at 25°C. The average optical linewidth of comb lines is measured as 440 kHz. Single-channel non-return-to-zero modulation rates of 70 Gbit/s and four-level pulse amplitude modulation of 40 GBaud/s are also demonstrated. To further extend the comb bandwidth, an array of QD-CPMLs driven at separate temperatures is proposed to achieve 36 nm optical bandwidth (containing 60 comb lines with 100 GHz mode spacing), capable of a total transmission capacity of 4.8 Tbit/s. The demonstrated results show the feasibility of using the QD-CPML as a desirable broadband comb source to build future large-bandwidth and power-efficient optical interconnects.

An ultrafast microring modulator (MRM) is fabricated and presented with Vπ·L of 0.825V·cm. A 240 Gb/s PAM-8 signal transmission over 2 km standard single-mode fiber (SSMF) is experimentally demonstrated. PN junction doping concentration is optimized, and the overall performance of the MRM is improved. Optical peaking is introduced to further extend the EO bandwidth from 52 to 110 GHz by detuning the input wavelength. A titanium nitride heater with 0.1 nm/mW tuning efficiency is implemented above the MRM to adjust the resonant wavelength. High bit rate modulations based on the high-performance and compact MRM are carried out. By adopting off-line signal processing in the transmitter and receiver side, 120 Gb/s NRZ, 220 Gb/s PAM-4, and 240 Gb/s PAM-8 are measured with the back-to-back bit error ratio (BER) of 5.5×10-4, 1.5×10-2, and 1.4×10-2, respectively. A BER with different received optical power and 2 km SSMF transmission is also investigated. The BER for 220 Gb/s PAM-4 and 240 Gb/s PAM-8 after 2 km SSMF transmission is calculated to be 1.7×10-2 and 1.5×10-2, which meet with the threshold of soft-decision forward-error correction, respectively.