Silicon photonics promises high-density, functional, and portable photonic integrated circuits (PICs), mirroring the success of electronic integrated circuits (EICs). While significant progress has been made in developing silicon photonics foundries capable of mass-producing modulators, photodetectors, and lasers, these PICs haven't met the stringent requirements of applications like microwave oscillators, atomic physics, and precision metrology. These applications demand ultralow-noise laser sources with narrow linewidths to suppress amplified-spontaneous-emission noise. Furthermore, these lasers require isolation from back-reflections from downstream components, usually achieved by incorporating bulky optical isolators that increase complexity and cost. The challenge lies in integrating these crucial components—high-performance lasers and isolators—directly onto the chip. While III-V materials are necessary for efficient optical gain, integrating the magnetic materials used in standard optical isolators into CMOS fabrication processes remains problematic. This research offers a novel solution: using ultrahigh-quality (Q) factor cavities to achieve both ultralow laser noise and reduced feedback sensitivity, eliminating the need for isolators entirely. This dual benefit stems from the inherent properties of these high-Q cavities, drastically simplifying PIC fabrication and avoiding the inclusion of magnetic materials in CMOS foundries.

Existing research highlights the need for heterogeneous integration of non-group-IV materials, particularly III-V materials for optical gain, in silicon photonics. Studies have explored various integration architectures for high-performance lasers, amplifiers, and isolators. The challenge of integrating magnetic materials for optical isolators into CMOS-compatible processes has been widely acknowledged. Previous work has investigated the use of ultrahigh-Q cavities to improve laser coherence and stability, but these often involve separate lasers and resonators requiring free-space or fiber coupling, resulting in less than optimal noise performance, especially in the critical 1 kHz to 100 kHz range for microwave and sensing applications. While some progress has been made in self-injection locking of lasers with on-chip resonators, the laser frequency noise remained relatively high. This research builds upon these previous efforts, aiming to achieve both ultralow noise and isolator-free operation within a single, integrated device.

The researchers developed a novel three-dimensional (3D) integration architecture to seamlessly integrate III-V-based lasers with ultralow-loss (ULL) silicon nitride (SiN) waveguides. SiN is chosen for its superior performance in metrology, sensing, and telecommunications. The high-temperature annealing required for achieving ULL in SiN waveguides, however, is incompatible with standard back-end-of-line semiconductor processes. To overcome this, the researchers employed a 3D design effectively separating the PIC into layers with distinct photonic functionalities: a III-V gain layer, a Si PIC layer, a SiN redistribution layer (RDL), and a SiN ULL layer. A key innovation is the 4.8 µm separation between the Si and ULL SiN layers, protecting the ULL SiN from subsequent processing steps. Interlayer transitions are achieved via evanescent coupling and waveguide geometry design, with the RDL controlling coupling between active and passive layers. This 3D design enables the integration of ultrahigh-Q resonators (intrinsic Q = 50 million) with high-performance III-V/Si distributed-feedback (DFB) lasers. This 3D approach allows for multiple overlapping but decoupled photonic layers, a significant advancement over previous heterogeneous integration methods. The fabrication process involves both monolithic and heterogeneous 3D integration techniques. The researchers leverage self-injection locking of InP/Si DFB lasers to thermally tunable SiN ultrahigh-Q resonators to generate ultralow-noise lasers. This involves precise tuning of laser wavelength, ring resonance, and forward/backward signal phases. The experiments involve characterizing laser coherence and feedback sensitivity through various measurements, including delayed self-heterodyne beat spectroscopy, optical spectrum analysis, and oscilloscope recordings. The feedback sensitivity is assessed by varying the downstream reflection strength and analyzing the laser coherence dynamics, while the generation of tunable microwave signals is achieved through heterodyne beating of two low-noise lasers on a photodetector.

The 3D integrated PIC demonstrated ultralow laser frequency noise, achieving approximately 250 Hz/√Hz at a 10 kHz offset and 2.3 Hz/√Hz at the white noise floor for the through port, with an even lower white noise floor (1.7 Hz/√Hz) and 5 Hz fundamental linewidth for the drop port. This represents the lowest laser frequency noise reported for a single-chip device. This low noise is attributed to the ultrahigh-Q resonator integrated on the same chip. The researchers also demonstrated significantly improved feedback insensitivity. For the free-running laser, regime II (linewidth governed by feedback phase) was reached at an on-chip feedback level of -41 dB, while for the self-injection locked (SIL) laser, the critical feedback level for maintaining stable operation (regime I) increased to -14 dB (through port) and >-10 dB (drop port), representing a substantial improvement. Even with an increased on-chip feedback of -6.9 dB to the SIL drop port, a stable and constant laser linewidth was observed. Finally, the researchers successfully demonstrated tunable microwave-frequency generation using heterodyne beating of two low-noise lasers, achieving a tuning range exceeding 375 GHz, with the generated microwave signal's phase noise remaining independent of the carrier frequency. The ability to generate tunable microwave frequencies in this manner shows great promise for low-noise millimeter-wave and terahertz generation.

The results demonstrate a significant advancement in silicon photonics, addressing the limitations of previous integrated laser sources. The 3D integration approach successfully achieves ultralow noise and high feedback tolerance, eliminating the need for bulky and costly optical isolators. This is crucial for enabling the integration of complex photonic systems on a single chip, opening up opportunities for miniaturization and cost reduction in various applications. The superior performance in terms of noise and feedback tolerance, compared to previously reported results, highlights the effectiveness of the 3D architecture and the self-injection locking mechanism. The demonstration of widely tunable microwave signal generation further showcases the versatility and potential of this technology for applications requiring highly coherent and stable microwave sources.

This research successfully demonstrated a novel 3D integration scheme for creating ultralow-noise, isolator-free lasers in silicon photonics. The integration of ultrahigh-Q resonators significantly reduces phase noise and enhances feedback tolerance, leading to superior performance compared to existing technologies. The demonstration of widely tunable microwave generation using these lasers opens new possibilities for integrated photonic systems. Future work could focus on further reducing the thermorefractive noise through resonator design optimization, exploring the integration of additional components like amplifiers and modulators, and investigating the integration of other materials compatible with the 3D architecture.

While the achieved laser frequency noise and feedback tolerance are exceptional, there are still potential limitations. The current study uses relatively compact resonators, and the laser frequency noise is limited by thermorefractive noise. Larger resonators or spiral resonators could further reduce this noise. Additionally, the microwave generation is currently limited by the bandwidth of the off-chip photodetector; integrating higher-bandwidth photodetectors directly onto the chip could extend the tunable frequency range. Future work should also explore the long-term stability of the integrated system and its robustness under various environmental conditions.