Integrated photonics is rapidly evolving from a complementary technology to mitigate electronic bottlenecks into an independent and pivotal discipline. This is propelled by its distinctive advantages, including broad analog bandwidth, high integration density, low power consumption, and cost-effective manufacturing [1–3]. The inherent strengths of photonics, such as ultra-wide bandwidth, ultrafast computational speed, and low transmission loss [4–7], offer significant benefits across diverse application domains, such as communication signal processing [8–12], deep learning technology [13–15], and quantum signal processing [16–20].

The development of industrial products generally follows two distinct paradigms. One is characterized by integration and specialization, focusing on highly efficient processing for one or a specific set of functions. The other is defined by modularity and generalization, aiming to adapt to a wide range of application scenarios and satisfy diverse performance requirements. These paradigms represent the dual dimensions of industrial product design: depth and breadth. In electronics, these paradigms are exemplified by application-specific integrated circuits (ASICs) and field-programmable gate arrays (FPGAs), which have become foundational technologies for the third and fourth industrial revolutions [21], respectively. In the realm of integrated photonics, most photonic chips proposed or designed are predominantly custom-tailored for specific application requirements, referred to as application-specific photonic integrated circuits (ASPICs) [22,23]. Similar to ASICs, ASPICs require extended periods for design, iteration, and validation, with development cycles spanning approximately 12–24 months [11,24–26]. Lengthy processing limits the innovation and optimization of the device. Consequently, the development of programmable photonic integrated circuits (PPICs), which serve as the photonic counterpart to FPGAs, has attracted increasing attention from researchers. Inspired by FPGAs, the programming and control of optical paths in PPICs are discretized into simple interconnected combinations of programmable unit cells (PUCs) [27–30]. On one hand, with advanced definitions and commands at the software layer, PPICs can significantly reduce the need for computer-based simulation of optical physical fields, thus enabling rapidly functional verification of specific integrated optical applications [31]. On the other hand, the functional reconfigurability of PPICs allows them to serve as interim solutions until corresponding ASPICs are brought into production.

Capable of supporting any user-defined route with an arbitrary beam splitting ratio, MZIs have emerged as the most widely utilized PUCs in PPICs [5,12,25–29,32–34], along with the advantages of wide operating bandwidth, low insertion loss, and high processing tolerance. However, the waveguides required for tunability in MZIs necessitate longer lengths to achieve lower tuning power, which limits their integration density. Several alternatives have been proposed to address the challenge, such as multimode interferometers [35], dual-drive directional couplers [36], multiport directional couplers [37], and periodic dual-mode waveguides [24]. Despite these efforts, these alternatives have yet to achieve sufficient compactness to replace MZIs. Additionally, their shrinking of the bandwidth limits their practicality. Moreover, when reconfigured as resonant devices, such as ring resonators and Sagnac ring resonators, MZI-based schemes suffer from a significantly limited free spectral range (FSR) due to the oversized resonant cavity. The limitation diminishes the practicality of MZI-based PPICs for wavelength-selective functions. In response, considerable research has also been devoted to PPICs based on micro-ring resonators (MRRs) [38,39] and micro-disk resonators (MDRs) [40], which realize wavelength-sensitive microwave photonic processing functions with extremely high space utilization.

In this work, we propose a novel PUC based on an MZI cascaded with parallel double-ring resonators (PDRRs), referred to as an MZI-cascaded-ring unit (MCR). The PDRR introduces additional tunable parameters, unlocking wavelength-selective functionalities. Furthermore, the small bending radius of the micro-rings ensures a large FSR, which facilitates the implementation of traditional MZI-based linear computing functions within the FSR. This significantly expands the functional coverage of the PPIC. To enhance integration density and reduce the power consumption required for tuning, spiral waveguides are employed in the place of straight waveguide interference arms. To validate the advantages of the proposed MCR-based programmable photonic processors, a PPIC with a circular array of four hexagonal meshes is fabricated and experimentally demonstrated to support a variety of optical processing functions. These functions cover tunable optical filters, Fano response resonators with tunable spectral resolution, multiparameter-tunable wavelength division multiplexing (WDM) systems, ring resonators, Mach–Zehnder interferometers, ring-assisted Mach–Zehnder interferometers, a four-channel optical router, tunable delay lines, and optical time-domain differentiators. In the future, this architecture is anticipated to enable additional functionalities such as beam forming, pulse shaping, and large-scale matrix operations for neural networks, further broadening its applicability in emerging photonic technologies.

A 3D schematic view of the proposed programmable photonic processor is illustrated in Fig. 1(a), employing a transversely compressed hexagon as the fundamental grid cell. Each edge within this cell represents the proposed MCR. To leverage the high refractive index contrast and ensure compatibility with existing complementary metal-oxide-semiconductor (CMOS) technology, silicon-on-insulator (SOI) is selected as the material platform for the device fabrication. The edges and ports of the proposed 4-HEX MCR-PPIC are systematically numbered for clear descriptions in Fig. 1(b). The MCR structure, depicted in Fig. 1(c), consists of a tunable coupler implemented via a balanced MZI for broadband optical power splitting, followed by a PDRR for single-wavelength or narrowband optical characterization. To verify the functionality and reconfigurability of the proposed MCR-based programmable photonic processor (hereinafter referred to as MCR-PPIC), a photonic processor chip with four hexagonal meshes (4-HEX MCR-PPIC) is fabricated and tested. The image of the chip used for testing and characterization is presented in Figs. 1(e) and 1(f), along with microscopic images. See Appendix A for more details about the processor fabrication and packaging.

A schematic of a typical balanced MZI-type tunable optical coupler is shown in Fig. 2(a). In this configuration, the phase difference between the two interference arms of the MZI is controlled through thermal tuning, resulting in corresponding changes in the outputs due to interference effects. The transmission matrix of the balanced MZI-type tunable coupler is given by Eq. (1) [41]: [Et1Et2]=ejΔφ−π2[sinΔφ2cosΔφ2cosΔφ2−sinΔφ2][Ei1Ei2],(1)where Δφ represents the phase difference between the two interference arms of the MZI. Δφ1 and Δφ2 represent the phase change due to thermal tuning of two waveguides, respectively. By ignoring the initial phase difference, Δφ=Δφ1−Δφ2. As shown in Fig. 2(b), spiral waveguides are employed as the interference arms. By testing the response of light transmittance to heating power, the power consumption for switching is approximately 16 mW/π, increasing by 34% compared with our previous work [42].

The transmission matrix of a single add-drop micro-ring resonator, as depicted in Fig. 2(c), is expressed in Eq. (2) [38,43]: [Et1Et2]=11−τ1*τ2*αejθ[−κ1*κ2αejθ2τ2−τ1*αejθτ1−τ2*αejθ−κ2*κ1αejθ2][Ei1Ei2],(2)where τ1 and τ2 are the self-coupling coefficients of coupling regions, respectively. κ1 and κ2 are the mutual-coupling coefficients. α denotes the loss factor within the micro-ring, where α=1 corresponds to a lossless ring. The phase accumulation θ describes the phase shift experienced by light as it circulates within the ring to reach a steady state, expressed as θ(λ)=2πλneff(λ)·Lr, where Lr is the circumference of the ring and neff is the effective refractive index of the transmission mode, which is wavelength-dependent. Under the condition of optical lossless coupling, the self-coupling and mutual-coupling coefficients within the same coupling system satisfy energy conservation relationship: τ2+κ2=1.

More ring resources can meet more flexible application scenarios, so a PDRR in Fig. 2(d) is actually used. Setting the phase shift to be introduced by the connecting waveguide between the rings as δ and the input signal vector as [Ei1Ei2]=[10], then the output signal Et1, Et2 shown in Fig. 2(d) can be represented as $$\begin{gathered} Et1=(τ11−τ12*α1ejθ11−τ11*τ12*α1ejθ1ejδ·−k21*k22α2ejθ221−τ21*τ22*α2ejθ2)ejδ+−k11*k12α1ejθ121−τ11*τ12*α1ejθ1, \\ Et2=τ11−τ12*α1ejθ11−τ11*t12*α1ejθ1ejδ·τ21−τ22*α2ejθ21−τ21*τ22*α2ejθ2. \end{gathered}$$(3)

To ensure that the input and output ports of the device are on the same side, respectively, a crossing waveguide is introduced to optimize the PDRR, as shown in Fig. 2(e).

In summary, the transmission matrix of a PDRR subcomponent can be expressed as [Et1Et2]=[h11h12h21h22][Ei1Ei2],(4)$$\begin{gathered} h11=τ11−τ12*αejθ1−τ11*τ12*αejθ·k21*k22αejθ2τ21*τ22*αejθ−1·ej2δ−k11*k12αejθ21−τ11*τ12*αejθ, \\ h12=τ11−τ12*αejθ1−τ11*τ12*αejθ·τ21−τ22*αejθ1−τ21*τ22*αejθ·ejδ, \\ h21=τ22−τ21*αejθ1−τ21*τ22*αejθ·τ12−τ11*αejθ1−τ11*τ12*αejθ·ejδ, \\ h22=τ22−τ21*αejθ1−τ21*τ22*αejθ·k12*k11αejθ2τ11*τ12*αejθ−1·ej2δ−k22*k21αejθ21−τ21*τ22*αejθ. \end{gathered}$$(5)

Therefore, according to Eqs. (1), (4), and (5), the transmission matrix of the MCR can be expressed as [Et1Et2]=[h11h12h21h22][sinΔφ2cosΔφ2cosΔφ2−sinΔφ2][Ei1Ei2].(6)

The specific dimensional parameters of the PUC are labeled in the accompanying figures. The overall size of the MCR structure measures 346μm×174μm. A spectral test experimental setup (see Appendix B for more details about the spectrum measurement) is used to characterize and measure the spectral response of the MCR in detail. The MCR operates in a wavelength-division-multiplexing manner, exhibiting distinct performance across different wavelength bands. When the input signal falls within the FSR of the PDRR, the spectral characteristics of the PDRR are obscured, allowing the PUC to be approximated as an MZI for linear computational tasks. Conversely, when the input signal lies outside the FSR, with MZIs in either bar or cross states, they can be simplified as straight waveguides. This allows the PUCs to exhibit the wavelength-dependent behavior of micro-ring resonators, supporting precise amplitude and phase manipulation of narrowband signals. Such functionality is particularly advantageous for signal processing tasks, such as microwave photonic filtering. More generally, due to the continuous tunability of the coupling coefficients, MCRs offer a significantly higher degree of reconfigurability compared to simple superposition of MZIs or MRRs. The results presented in Appendix C confirm the superiority of MCRs as fundamental building blocks, establishing a robust foundation for the development of advanced programmable photonic processors.

In order to verify the reconfigurability of the proposed processor, different experimental test links have been built to test its functionality in various aspects. The specific experimental links are shown in Appendix B. The PUCs of the chip were fully calibrated prior to usage. In this section, only the MZI subcomponents of the MCR are utilized to demonstrate the functional reconfiguration of the processor, shown in Fig. 3. Balanced Mach–Zehnder interferometers (BMZIs) function as optical switches, enabling the routing of optical signals between different paths. The optical path demonstration and experimental results for a BMZI are illustrated in Fig. 3(a). The spectra exhibit the broadband characteristics of the BMZI within the wavelength range of 1550–1555 nm. Switching states are achieved by applying different 3 dB voltages to Edge-D, resulting in phase differences of either 0 or π between the outputs of Edge-D. In Fig. 3(a-ii), the color blocks are the regions surrounded by the upper and lower envelopes of the spectral lines, and the solid line is the result after smoothing processing. Unbalanced Mach–Zehnder interferometers (UMZIs) serve as fundamental components in lattice and finite impulse response (FIR) transversal filters, which are widely used in linear phase filters, biosensors, and group delay compensators. As shown in Fig. 3(b), UMZI is configured with a path length difference of 4×BUL (basic unit length). The extinction ratio can be controlled by adjusting the voltages applied to Edge-D and Edge-Q without impacting the FSR. In Fig. 3(c), a UMZI with a path length difference of 6×BUL is demonstrated by adding Edge-J. By comparing the FSR for different path length differences, the result shows that the FSR for a 4×BUL difference is about 0.1 nm, while for a 6×BUL difference, it is around 0.067 nm. The inverse proportionality between the FSR and the path length difference is consistent with theoretical predictions.

Ring resonators are versatile components that can be configured in both all-pass and add-drop configurations, each offering distinct filtering capabilities. The all-pass configuration can implement all-pole infinite impulse response (IIR) notch filters, while the add-drop can realize both IIR notch and FIR+IIR bandpass filters. An all-pass ring resonator with a cavity length of 6×BUL is depicted in Fig. 3(d). By adjusting the voltage on Edge-A, different resonance depths of the resonator can be achieved. When the voltages on Edge-A are 1.36 V and 2.34 V, the spectral fluctuations are minimal, indicating that the processor is operating near the 1×BUL and 6×BUL straight waveguides. A significant change in resonance depths is observed when the voltage varies between these two values. In Fig. 3(e), by setting Edge-B to be tunable as well, the processor can be configured as an add-drop ring resonator. The response of the drop port becomes progressively more pronounced at different voltage settings. The peak output signals from the through and drop ports essentially aligned at the voltage group of [2.0 V, 1.7 V]. The optical paths and spectra of all-pass ring resonators with different ring lengths are demonstrated in Fig. 3(f). The inverse relationship between the FSR and the cavity length is in agreement with theoretical predictions. Specifically, the device proposes two alternative optical paths for reconfiguring the resonator with 12×BUL cavity length. Both configurations exhibit comparable performance, underscoring the flexibility and robustness of the processor’s reconfiguration capabilities. This approach ensures that specific functions can still be executed via alternative optical paths even if a failure occurs in a particular MCR unit.

The implementation of more complex filters using the MCR-based programmable photonic processor is shown in Figs. 3(g) and 3(h). As depicted in Fig. 3(g), the coupling coefficient of the series double-ring filter between rings is controlled by Edge-J. At Edge-J voltages of 1 V and 1.9 V, the spectra exhibit nearly periodic characteristics. However, at other voltages, the spectra become heterodyne, which can be used to broaden the filter’s bandwidth and introduce controllable ripples. As shown in Fig. 3(h), the coupling coefficient between the MZI interference arm and the ring in the single-ring assisted MZI filter is controlled by Edge-J. At a voltage of 1.7 V, the filter exhibits a nearly flat-top characteristic. The filter can serve as a fundamental building block for special configurations, including high-order Butterworth and Chebyshev filters.

An optical router is a critical component in optical communication networks, primarily utilized for the efficient transmission and switching of optical signals. To ensure stable data exchange in the network, it is essential to accurately route optical signals from input ports to target output ports according to network requirements. In Fig. 3(i), a 4×4 optical router is obtained by using the MCR-PPIC, enabling information exchange between any left and right ports. The PUC highlighted in orange in Fig. 3(i-i) can be configured in either the bar state or cross state depending on the actual requirements. As shown in Fig. 3(i-ii), 16 transmission spectra between the left and right ports exhibit crosstalk levels below −15dB in all cases, while the main output response remains flat within the bandwidth, with an insertion loss of about 12 dB. The insertion loss is slightly larger. On the one hand, the end-face coupling loss is about 8 dB. On the other hand, basic devices such as MMI need further optimization. Thus, the work can have sufficient practical value in the future. In addition, the performance characterization can be further optimized by complementing the spectral self-configuration system. In fact, the proposed optical router can be regarded as a 4×4 optical matrix, where the elements are either 1 or 0, representing the presence or absence of a connection between input and output ports. Depending on the operating characteristics of the MCR-PPIC, it is feasible to reconfigure this matrix into any arbitrary unitary matrix during the linear computations of the optical neural network. Moreover, the inclusion of PDRR subcomponents further enhances the flexible control over both the amplitude and phase of the matrix elements. This capability theoretically allows the processor to realize arbitrary matrices, paving the way for advanced applications in optical signal processing and neural network computations.

A time delay test experimental setup (see Appendix B for more details about the time delay measurement) and the delay measurements utilizing the MZI subcomponents in the MCR structure are shown in Fig. 4. The optical path begins at Edge-I and sequentially incorporates Edge-LP, RS, and QO. In Fig. 4(a), the 1547.4–1549 nm band in the spectra corresponds to the micro-ring resonance region, which is initially hidden in the figure to focus on the broad spectral characteristics. The delay results, presented in Fig. 4(b), have subtracted the delay contribution of the outer link. The delay spectra exhibit a flat response within the wavelength range from 1545 to 1555 nm. Furthermore, a significant increase in delay is observed as the number of BULs increases, from 147.12 ps in a single stage to 265.79 ps in 7-stage cascade, validating the device’s capability for discrete delay tuning.

The PDRR subcomponents within the MCR excel at executing wavelength-selective spectral functions and exhibit excellent complementary properties with the reconfigurable capabilities of the MZI subcomponents. The realization of various microwave photonic filters using the PDRR subcomponents is demonstrated in Fig. 5. The corresponding optical path schematics are embedded within each spectral result. Since the MZI subcomponents operate exclusively in either the cross or bar state for these functions, functionally transparent to the PDRR, they have been omitted from the schematics. The serial numbers of the edges involved in the functional reconstruction are labeled in the figures. As illustrated in Figs. 5(a) and 5(b), the bandpass filter can be reconstructed by the rings of Edge-C and Edge-E. The 3 dB bandwidth of the single Ring-C1 is 0.177 nm, as shown in Fig. 5(a). By simultaneously activating both rings of Edge-C and controlling their resonance wavelengths, the bandwidth of the bandpass filter can be effectively extended, resulting in an excellent flat-top response. By activating Ring-C2 and tuning its center wavelength, the bandwidth of the reconfigured flat-top filter increases from 0.230 to 0.285 nm, and the in-band ripple is kept within 0.2 dB. Compared to the single-ring configuration, the flat-top filter exhibits steeper stopband attenuation. By activating the ring within Edge-E and utilizing its characteristic of power suppression at the through port, the spectral shape of Ring-C1 can be modified to reduce its bandwidth, better adapting to the requirements of ultra-narrowband filtering in microwave photonics. Meanwhile, the stopband attenuation of the response is enhanced, bringing the overall spectrum closer to that of an ideal box filter. The tunability of the center wavelength of the flat-top filter is illustrated in Fig. 5(b), showing a red shift of 1.05 nm when the voltage increases to 1.7 V.

The experimental results of the reconfiguration of the bandstop filter with tunable parameters, including the stopband depth, stopband width, and center wavelength, are presented in Figs. 5(c)–5(e). In Fig. 5(c), by tuning the stopband width while maintaining the stopband depth greater than −30dB, the 10 dB stopband width increases from 57 pm for a single ring to 240 pm for four rings, representing an approximately threefold increase. This demonstrates the capability for continuous and wide-range tuning of the bandwidth. In addition, while keeping the 10 dB stopband width constant, the center wavelength shifts by 1.39 nm during tuning, as illustrated in Fig. 5(d). In Fig. 5(e), by aligning the center wavelengths of different ring resonators, the stopband depth can be significantly increased from −27.35 to −63.16dB.

By adjusting the resonance wavelengths of Ring-C1 and Ring-E1, the spectral line can exhibit a Fano line shape feature, with a tunable spectral resolution and slope direction, as shown in Fig. 5(f). According to Fig. 5(f), it can be observed that the notch shows a clear redshift tendency under the thermal tuning effect and passes over the original bandpass peak. Therefore, it can be inferred that an electromagnetic-induced transparent (EIT)-like spectrum can be constructed under appropriate tuning power on Ring-E1. These results highlight the significant potential of PDRRs in reconstructing microwave-photonic functions.

The results of group delay measurements utilizing the PDRR subcomponents are presented in Figs. 5(g) and 5(h). The delay characteristics of the PDRR components are demonstrated in Fig. 5(h). Based on the delay properties of an MZI subcomponent, a single ring can provide continuous delay variations ranging from +20 to −12ps, showcasing its flexible delay tuning capability. In addition, when the two rings are configured in a certain combination, a flat delay region is observed within the range of 1548.19–1548.30 nm (corresponding to a bandwidth of 10 GHz). The region provides a consistent delay amount of approximately +10ps, which can be effectively utilized in future applications requiring stable and predictable delay profiles. Flexible delay tuning capabilities make it well-suited for beamforming as well as optical phased arrays.

Wavelength division multiplexing (WDM) systems significantly enhance transmission capacity by simultaneously transmitting multiple optical signals at different wavelengths through a single physical medium. These systems are widely used in long-haul trunks, metropolitan area networks (MANs), and data center interconnections. A WDM system featuring scalable wavelength channels, a tunable channel bandwidth, and channel spacing can be dynamically reconfigured using the proposed 4-HEX MCR-PPIC. The maximum number of channels (Nch) that can be supported is determined by Nch=FSRChannel Spacing. The reconfigured optical path is shown in Fig. 6(a), where broadband light enters via Port-R1, and narrowband spectra are output from Ports-L1, L6, L3, and L5, respectively, achieved through the operation of micro-rings within the four edges (Edge-IECA). As illustrated in Fig. 6(b), the channel scalability of the device shows the incremental increase in the number of channels from one to four. The crosstalk in Port-L6 is the highest among all channels at 18.2 dB, which remains within an acceptable range for practical application. The tunability of the channel spacing is verified in Fig. 6(c), where the spacing increases from 0.49 to 0.75 nm. This large tunable range highlights the flexibility of the MCR-PPIC in adapting to different WDM system requirements. In Fig. 6(d), the tunability of the channel bandwidth is validated. Particularly, by aligning the resonance wavelength to the Port-L1 or away from the current operating channels, an extremely narrowband channel can be realized with a 3 dB bandwidth of 0.176 nm. Additionally, the in-band ripples of the channels are less than 1 dB (typically between 0.8 and 0.9 dB), allowing the 3 dB bandwidth to reach about 0.4 nm. This performance meets the requirements of current DWDM systems and holds practical significance.

Furthermore, the capability of the micro-ring resonator to process time-domain signals is evaluated using experimentally measured delay and spectral data in conjunction with simulation tools. The phase-frequency response of the micro-rings is inversely deduced through numerical integration based on the relationship τ=dϕ/dω, where τ is the group delay, ϕ is the phase, and ω is the angular frequency. The amplitude-frequency response is derived from spectral analysis. Both responses are imported into the Lumerical INTERCONNECT simulation software for optical time-domain signal differentiation experiments. As shown in Fig. 7, the differentiation experiments are conducted separately for each of the two rings (Ring-I1 and Ring-I2). The original input signal is a Gaussian pulse, and the output signals exhibit two distinct peaks, confirming the successful execution of the differentiation calculations. The differential signals in Figs. 7(a) and 7(d) exhibit a noticeable time delay compared to the initial input signals, with the time domain waveforms broadening. This phenomenon may be attributable to the dispersion effect of the transmission medium. By normalizing the amplitude and compressing the width of the differential signals, and comparing them with the theoretically calculated signals, it is evident that the differential order executed by Ring-I1 is 0.5, while that of Ring-I2 is 1.15. Both values match well with the theoretical predictions. Combined with Fig. 5(h), the two micro-rings exhibit similar delay characteristics but differ in their differential orders. Figures 7(c) and 7(f) represent differentiator spectra and signal spectra, and a significant suppression of the spectrum at resonance wavelength can be observed. This suggests that the differential order of the micro-rings can be controlled by tuning the operating wavelength. The experiment demonstrates that the two micro-rings operate independently in terms of differentiation, proving that the MCR-PPIC is capable of differentiating multiple optical signals simultaneously in both the spatial and frequency domains, which greatly enhances the efficiency of signal processing.

The experimental results presented above highlight the potential of the proposed MCR and the 4-HEX MCR-PPIC. For the MCR, it cascades an MZI tunable coupler with a PDRR optical switch, combining wavelength selectivity with a large FSR while retaining the original broadband optical processing capabilities. Compared to the traditional MZI or MDR unit, this novel PUC provides enhanced spectral reconfiguration potential, significantly expanding the functionality and performance tunability of PPICs. To compensate for the increased lateral distance caused by the cascaded PDRR subcomponents, spiral waveguides are implemented in place of straight waveguides in the MZI interference arms. This design achieves a final lateral length of 346 μm, representing an improvement over previous works [27,28]. In addition, the introduction of the spiral waveguides optimizes the thermal tuning units, reducing the power consumption required for the same phase shift and enhancing the energy efficiency of the processor. Furthermore, the increased spacing between thermal electrodes along the longitudinal distance mitigates the adverse effects of thermal crosstalk. After testing the characteristics, the extinction ratio of the MZI subcomponent in the proposed MCR is over 20 dB within the FSR. The time delay is approximately 14 ps, and the tuning efficiency is around 16 mW/π. The extinction ratio of the PDRR subcomponent is over 25 dB, and the time delay ranges from +20 to −12ps. The dynamic characterization of the PDRR is detailed in Appendix A. Based on the MCR programmable unit, a programmable photonic processor with a 4-hexagonal-mesh architecture is designed, fabricated, and characterized. This design demonstrates the processor’s versatile configurability in optical paths as well as capability to perform various microwave photonics-related functions. These functions include configurable balanced and unbalanced FIR Mach–Zehnder filters, ring cavities with discretely variable ring lengths, CROW structures, ring-assisted MZI filters, discrete tunable delay lines, and a 4×4 optical router, all utilizing the MZI subcomponents of the MCR. Moreover, the PDRR subcomponents are also utilized to configure various tunable optical filters and systems. These include bandwidth- and wavelength-tunable bandpass filters, depth-, bandwidth-, and wavelength-tunable bandstop filters, spectral-resolution-tunable Fano line filters, a continuous tunable delay system, a multi-channel time-domain differentiator with adjustable differential order, channel numbers, bandwidths, and spacing tunable WDM systems. Theoretically, the proposed MCR-PPIC can realize all the functions of MZI- and MDR-PPICs reported to date, verifying its strong versatility and reconfigurability. In addition, limited by the experimental conditions and other factors, this paper has yet explored the application potential of the device, especially the linkage between the MZI and PDRR subcomponents. This is something to look forward to, such as combining the discrete delay of the MZI with the continuous delay of the PDRR to achieve a larger range of continuous delay tuning. A comparison between previously reported programmable photonic processors and our work is shown in Table 1. Compared with the AMZI-MRR unit mentioned in Table 1, the proposed MCR has more ring resources participating in spectral reconstruction. Moreover, the participation of the MZI enables the FSR band to be fully utilized as well, which gives it obvious reconstructive advantages. On the other hand, the MCR can be extended flexibly, and its processor application potential is stronger after extension. Regarding the PBW unit, the MCR also has the advantage of a large operating bandwidth.Table 1.

Comparison of Reported Programmable Photonic Circuits on SOI

Currently, our work still has significant room for improvement. In practice, the limited availability of current sources required to tune the MCR restricts the number of configurations we can demonstrate. Our reconfiguration demonstrations are divided into two main groups: the MZI subcomponents and the PDRR subcomponents. However, the two subcomponents are not mutually exclusive. As mentioned, their combination in delay line function can achieve a wide range of continuously tunable delays. Additionally, during our design process, we observed that in current optoelectronic hybrid integration, the size of thermal electrodes is approximately two orders of magnitude larger than that of optical devices, which limits further integration of the components. For certain functionalities, managing thermal crosstalk in PUCs is crucial for practical operations. Recent studies [44–46] have proposed solutions to mitigate thermal crosstalk in CMOS-compatible silicon photonic platforms. Adopting low-crosstalk tuning methods to replace thermal tuning is also a promising direction for future research. Further, fabrication errors introduce uncertainty in the initial state of MZI subcomponents within the MCR, which affects the practicality of the reconfigurable photonic processor. Current research efforts focus on calibration-free MZIs [47,48] and alternative reconfigurable unit schemes such as PBWs [24]. As the scale of expansion increases, developing more effective control algorithms and software for photonic processors becomes increasingly important and is a critical area for future research [25,26,49].

In summary, in this work, we designed and experimentally demonstrated a highly compact and versatile programmable photonic processor based on reconfigurable MCRs. Compared to existing MZI-based solutions, the proposed device achieves a reduced footprint, lower power consumption for controlling the optical units, and wider functional coverage and versatility, signifying a further advancement toward large-scale, high-density programmable photonic processors. The proposed photonic processor has broad potential applications in microwave photonic communications, biophotonics, sensing, multi-processor interconnects, switching, and quantum information, representing a significant step toward a new paradigm for highly versatile programmable photonic processors.