1. Introduction

With the rapid development in the field of artificial intelligence, although it is possible to use GPUs and ASICs to improve computing efficiency, the limitations in terms of energy consumption and heat dissipation have become a challenge due to the bottleneck of the von Neumann architecture. This has led researchers to urgently seek alternative computational methods that can break the traditional von Neumann architecture-based computer interconnectivity. Silicon-based photonics has gained increasing attention in the scientific community. Photonic integrated circuits not only offer high information transmission rates and bandwidth but also overcome the limitations imposed by traditional metal interconnects. The significant advantages of photonic circuits in signal transmission have garnered considerable interest. In particular, non-volatile PCMs have been widely applied in photonic devices such as optical switches [1,2], photonic multi-level memory [3,4], and absorbers [5] due to their unique optical contrast between different states and their zero static power consumption in maintaining states. The switching between amorphous and crystalline phases of PCMs induces a substantial change in refractive index, thereby modifying the properties of these devices which can operate between two stable states, “0” and “1”, without requiring additional energy to maintain the states [6]. Moreover, by adjusting the crystalline fraction of the PCMs, intensity variations in the output of waveguide devices can be achieved, enabling multi-level storage beyond “0” and “1”. This greatly enhances the storage efficiency and density of memory devices, providing more scalability for applications such as optical computing [3] and photonic synapses [7].

In recent years, the research of integrated phase change photonic devices has made remarkable breakthroughs. Reconfigurable devices with various structures were designed by combining waveguides and PCMs. For example, the combination of the Mach-Zender interferometer (MZI) and the PCM constitutes a compact standard multiplication unit with low insertion loss, which takes advantage of the optical properties of the PCM in different states [8,9]. Although MZI-based photonic neural networks perform well in terms of energy and speed, however, due to their relatively large size, the scalability of MZI is somewhat limited. Additionally, microring [10] and racetrack resonators [11,12] are also frequently used as optical switches and memory devices. Although both MZI and resonator-based devices can effectively control the transmission of light by modulating the crystallization fraction of the PCM, they have the disadvantage of large footprint, high energy consumption, and slow switching due to the large size of the PCM.

To mitigate the footprint and enhance the switching speed of PCMs, previous studies (Ref. [13,14]) have localized PCM within a metallic gap, coupling light through a conical Si₃N₄ waveguide to form a plasmonic metal-PCM-metal waveguide. This configuration allows for optical transmission to be modulated by the refractive index changes in the PCM across different crystallization levels. The approach leverages the localized heat generated by plasmonic excitation for rapid state transitions of the PCM, thereby achieving high energy efficiency and paving the way for low-power optoelectronic devices suitable for neuromorphic and in-memory computing applications. However, this method encounters significant optical transmission losses due to the slit structure, leading to inadequate optical contrast (< 15%). In pursuit of diminishing light transmission losses within the slit waveguide, a novel nano-antenna design was introduced in Ref. [15]. This design uniquely positions Ge₂Sb₂Te₅ (GST) within the gap between two metallic silver discs, exploiting the plasmonic resonance induced by the interaction between light and the metal to intensify the electric field at the GST site. Such enhancement significantly boosts the light modulation effect of the GST, culminating in a notable optical contrast of 12.8% while still maintaining a minimal erase/write energy requirement of 15/2 pJ. The tunable nanophotonic device was proposed by using photonic crystal cavities embedded with phase-change materials. By exploiting the strong interaction between the resonant mode of the cavity and the phase-change material, a high contrast of around 14 dB was achieved [16]. This advancement marks a significant step forward in the development of efficient and high-contrast optical modulation mechanisms for next-generation optoelectronic applications.

To enhance optical contrast, researchers have opted to replace the single-mode Si₃N₄ waveguide with a more compact Si waveguide, characterized by tighter light confinement and improved propagation control. This modification facilitates efficient light transmission with reduced scattering and improved optical contrast. Furthermore, a plasma nanoantenna is created by embedding GST into a metal Ag ring. This innovative design induces a plasma resonance at the top of the Si waveguide, effectively enhancing the modulation effect of the GST on light. The plasma resonance enables precise control over the phase change process, leading to a remarkable optical contrast of 28.5%. Additionally, the utilization of light heating compensates for the inherent slow response of electric heating in PCMs [17]. Though the plasma effect can enhance the modulation of light by PCMs, plasma devices face limitations due to the exponential decay of the plasma-enhanced electric field region at the dielectric junction. Achieving high contrast is crucial as it allows for more precise transmission and detection of optical signals and enables higher storage density.

To overcome the drawbacks of existing devices characterized by low optical contrast and high switching energy, a novel switching structure is proposed by combining the PCM-hybrid nanoantenna with the F-P cavity. The metal nanoantenna is deposited on an F-P cavity Si waveguide composed of symmetric gratings. PCMS is placed inside the nanoantennas. For the selection of PCMs, GST is a commonly used chalcogenide compound for wide applications, from phase-change optical discs to current phase-change photonic memories due to the significant changes in optical properties between the crystalline and amorphous phases in the near-infrared wavelength range [18]. Compared with other phase-change materials such as Sb₂S₃ [19], Sb₂Se₃ [19], Ge₂Sb₂Se₄Te [20], and Sb [21], GST stands out due to its substantial refractive index disparity between its crystalline and amorphous states. This characteristic is pivotal for enhancing light modulation efficiency and facilitating superior performance under the constraint of reduced phase-change material dimensions. The interaction between the input light and the GST is enhanced using F-P cavity resonance and nanoscale metal plasmon resonance effects [22,23]. This effectively confines the waveguide mode within the GST, allowing for faster switching speed, lower optical loss, and higher optical contrast in smaller structure sizes [24]. This device operates in a fully integrated manner, where readout and switching pulses are guided through on-chip optical paths. This makes it easier to interconnect with other photonic devices to build complex on-chip photonic networks with all-optical capabilities, enabling integrated all-optical non-volatile devices. At a communication wavelength of 1550 nm, it achieves large switching contrast and low energy consumption with ultra-small footprints. It also realizes more accurate multi-level storage capabilities.

2. Device structure and working principle

The proposed device structure incorporates an F-P resonance by introducing a phase-shift cavity containing GST in the center of the grating. This arrangement leads to resonant transmission peaks within the bandgap of the photonic structure. When light reaches the cavity, it undergoes forward and backward reflections, resulting in the formation of standing waves at the resonant wavelength. As a consequence, the electric field is significantly enhanced at the central position of the resonant cavity. Placing the nanoantennas at the center position of the standing waves allows for the coupling of light with the PCMs Additionally, the combination of the PCMs with the nanoantennas utilizes the plasmonic effect of Ag to further enhance the interaction between the PCMs and light.

The RF module in COMSOL was employed to calculate the localized field induced by plasmas and the device's emission power by solving the classical Maxwell's equations. To avoid any non-physical reflections, we applied the second-order boundary scattering condition to the outer surface. During the simulation of the phase-change process, we utilized the COMSOL heat transfer (HT) module for numerical simulations. It is noted that the presence of surface roughness or lattice defects (such as lattice mismatch, dangling bonds, impurities, missing atoms, and substitutions) can cause substantial deviations from the anticipated thermal behavior. The thermal boundary resistance (TBR) is introduced to impose a limit on the heat flux across the interface [25]. We calculate the temperature inside the GST cell by solving the heat conduction equation and then use the temperature in the rate equation to calculate the degree of phase transition. The rate equation is deduced from the MRE model proposed by S. Senkader and C. D. Wright [26]. Simultaneously solving the coupled equations enables direct prediction of the crystallization degree of the GST layer, achieving amorphization at a temperature of 893 K, followed by rapid cooling. All material settings and parameters are provided in Table 1.

Table 1. Material parameters for the device simulation.

View Table | View all tables in this article

Figure 1(a) illustrates the structure of a resonant optical memory. It consists of a straight waveguide inserted into an F-P cavity. The two ends of the F-P cavity are single-mode Si waveguides with cross-sectional dimensions of 220 nm × 500 nm. The F-P cavity is formed by sidewall gratings acting as the front and back mirrors, with a total of N = 20 grating periods. At the top center of the Si waveguide, there is a nanoscale antenna with PCM surrounded by a metal ring made of Ag. Figures 1(b) and 1(c) depict the structure of the nanoscale antenna. The PCM has a thickness of 20 nm, and it is enclosed by a silver ring with a width of W_Ag= 80 nm. Figure 1(d) displays the structure of the Bragg grating reflector. We first set the Bragg resonant wavelength to be 1550 nm. The Bragg resonance condition is solved by the following formula:

 

Fig. 1. Schematic diagram of the proposed F-P cavity plasmonic memory. (a) A three-dimensional perspective of the memory, consisting of symmetric F-P resonators and a PCM surrounded by a plasma nanostructure of elliptical silver rings. (b) Top view and (c) cross-section of the phase change region. (d) Top view of the Bragg grating mirror.

Download Full Size | PDF

Figure 2(a) illustrates the reflection spectrum of the grating device obtained through simulations using the three-dimensional finite-difference time-domain method (3D-FDTD). It can be observed that when the grating period Λ is 337 nm and the duty cycle is 0.5, the grating exhibits a strong reflection at the incident wavelength of λ = 1550 nm. Figure 2(b) shows the relationship between the F-P cavity waveguide and the incident wavelength in the absence of load. It can be seen that the highest transmission occurs at 1550 nm, and there is a clear resonance phenomenon. Figure 2(c) depicts the spatial distribution of electric field intensity under two distinct modes: one characterized by a low transmission wavelength of 1535 nm, and the other by a high transmission wavelength of 1550 nm. Notably, the intensity at the output port corresponding to 1535 nm is markedly lower compared to the input port. This discrepancy strongly suggests that a substantial portion of the incident light is reflected at 1535 nm. Consequently, the transmission efficiency at this lower wavelength seems to be significantly hindered or attenuated within the device. The F-P cavity effectively realizes the filtering ability, and at the wavelength of 1550 nm, the electric field in the center of the waveguide is significantly enhanced, which provides a promotion effect for the subsequent interaction between light and PCM.

 

Fig. 2. (a) Simulated reflection spectrum of a Bragg grating with N = 20 periods. (b) Transmission spectra of an unloaded F-P resonant cavity. (c) Top-down view of electric field profiles of two modes observed at incident wavelengths λ = 1535 nm and 1550 nm.

Download Full Size | PDF

Plasmonics exploit the coupling between light and the collective electronic charge oscillations in metals, localizing light to sub-wavelength dimensions and confining the electric field near the nanoscale metal surface. This results in stronger local electric field enhancement, amplifying the phase-change characteristics of the PCM. In the near-infrared wavelength range, silver (Ag) exhibits excellent plasmonic properties [35]. Placing the PCMs within an elliptical Ag ring reduces the overall footprint of the antenna. The high refractive index of the waveguide confines light to diffraction-limited optical modes. When plasmonic resonance occurs, the field enhancement within the gap significantly enhances the coupling between light and the PCMs and enables faster switching of the PCMs. Additionally, this hybrid structure can promote uniform phase change of the PCM under optical pulses, avoiding non-uniform crystallization. However, the diffusion effect of Ag ions can cause changes in the optical properties of sulfur-based compounds. Therefore, the addition of appropriate barrier layers such as Si₃N₄ and TiN is necessary to effectively prevent the diffusion of metals into the phase change materials to achieve high cyclicity of the memory [36].

 

Fig. 3. (a) Schematic diagram of geometric size optimization of nanoantennas. (b) The optical contrast mapping with the variations of the axial lengths r_a and r_b of the PCM. (c) The optical contrast mapping with the variations of the width and the thickness of the Ag ring. (d) The memory electric field distribution of amorphous and crystalline at a wavelength of 1550 nm. The cross-sectional distribution of the electric field in (e) amorphous and (f) crystalline states under an input power of 2 mW.

Download Full Size | PDF

Due to the complexity of its antenna structure, further optimization is required for the geometric parameters, including the lengths and thicknesses of the GST thin film and the width of the Ag ring. Different states of the embedded GST in the nanoscale antenna result in varying light absorption and transmission characteristics. To obtain the device with maximum contrast and a more uniform electric field distribution, numerical evaluation is performed during the optimization process, using signal contrast as a metric (signal contrast is typically defined as Ref. [11]: ${\textrm{Contrast}\; = \; }\frac{{{{T}_{{\textrm{am}\; }}}{ - \; }{{T}_{\textrm{cry}}}}}{{{{T}_{\textrm{am}}}}}$). Figure 3(a) shows the geometric information of the nanoantenna. The key factors affecting the performance of the nanoantenna are the long axis r_a and the short axis r_b of GST. To obtain a good trade-off between insertion loss and plasmonic effect, r_a and r_b were scanned in the 30-80 nm range. The size with r_a = 60 nm and r_b = 30 nm not only occupies a smaller footprint but also exhibits better contrast, as shown in Fig. 3(b). The width of the Ag ring and the thickness of the nanoscale antenna also affect the optical contrast. Figure 3(c) demonstrates that the contrast varies with the width of the silver ring and the thickness of the antenna, showing an increasing and then decreasing trend. A silver ring width of W_Ag = 80 nm and a nanoscale antenna thickness of 20 nm were chosen, resulting in a maximum contrast of 70.6%. Through the aforementioned size optimization, the structure of the entire device unit was determined.

Plasmonics can confine the electric field to a small region, and the difference in the dielectric constants of the PCMs in different states leads to variations in the electric field distribution. The electric field distribution of the device in crystalline and amorphous states is shown in Fig. 3(d). It is evident that in the region where GST is situated, there is a strong enhancement phenomenon observed in the local electric field. This pronounced and uniform enhancement of the electric field facilitates a homogeneous and rapid phase transition of the GST. As a result, the temperature difference in the whole PCMs region is reduced, preventing localized regions from reaching excessively high temperatures. The electric field distribution of the cross-sectional component is shown in Fig. 3(e) and (f). The mode field is locked in the resonance region (close to Ag) in the amorphous state. The light is also confined to the GST region in the crystalline state.

3. Results and discussion

3.1 Memory performance

This structure utilizes F-P cavity resonance and plasmonic effect to enhance the interaction between the incident light and the PCMs deposited inside the silver ring. This plays a crucial role in improving the performance of the device, particularly in terms of enhancing the switching speed and reducing energy consumption in non-volatile memories, especially for multi-level storage. When the incident light interacts with the PCMs, the transition of the material's state causes a drastic change in the local refractive index, resulting in a variation in the device's transmission. Due to the high refractive index and significant absorption of the crystalline GST, the transmission in the crystalline state is lower than that in the amorphous state. To intuitively observe the enhancement effects of these two phenomena on the devices (see Fig. 4(a)), the spectral transmission was compared when both the F-P cavity resonance and plasmonic resonance were acting together and when each factor was acting alone, with the same size of GST used, as shown in Fig. 4(b). When the incident wavelength is 1550 nm, using only F-P cavity resonance enhancement (green line) results in limited modulation of the optical field due to the ultra-small volume of the PCMs, resulting in only 1% optical contrast (see Fig. 4(c)). However, when plasmonic resonance is employed by using the Ag silver ring on the straight waveguide, significantly increasing the interaction between light and the PCMs, the optical contrast reaches 26%, as shown in Fig. 4(b). Moreover, when both F-P cavity resonance and plasmonic effect are combined, the optical contrast reaches 70.6%, which is around 69 times higher than when only F-P cavity resonance is used. The variation in insertion loss was also compared, as shown in Fig. 4(d). The insertion loss (IL) of the phase-change device is defined as IL = -10 log₁₀ (P_out/P_in), where P_in and P_out are the input and output powers, respectively. The overall device exhibits an insertion loss of only about 0.85 dB at 1550 nm wavelength in the amorphous state.

 

Fig. 4. (a) Schematic diagrams of the three device structures. Performance comparison: (b) transmission spectrum and (c) optical contrast change with wavelength. (d) Insertion losses in both crystalline and amorphous states.

Download Full Size | PDF

3.2 Phase-change process

The plasmonic effect, characterized by the confinement and enhancement of the electric field, facilitates the rapid attainment of phase transition temperature in PCMs when illuminated by optical pulses. The volumetric heat source (Q_V) is caused by the optical losses from the GST cell and the nanoscale antenna itself. The heat transfer equation is given by [25]:

The power consumption of the device is determined by the intensity and duration of the pulses required for write/erase operations. To distinguish it from the read signal (λ = 1550 nm), we set the wavelength for write/erase operations to 1549 nm. Figure 5 illustrates the temperature distribution of the PCMs during the crystallization and amorphization processes. The amorphization process of the PCMs is achieved by applying a pulse with a power of 0.6 mW and a duration of 1.5 ns. The maximum temperature of the PCMs (red line) rapidly rises to 1200 K, which is between the melting temperature (T_m = 893 K) and the temperature threshold for damage (1500 K). When the excitation pulse ceases, the temperature rapidly decreases, meeting the cooling rate requirement (within tens of degrees per nanosecond) for successful amorphization of GST [37]. In the low-refractive-index amorphous state, most of the electric field can still be confined within the PCM due to the plasma effect brought about by the silver ring surrounding the GST. Therefore, even during the crystallization process, it replaces traditional high-power pulses, reducing energy consumption. During the crystallization process, we apply an optical pulse of 0.5 mW - 10 ns to the amorphous GST, raising its temperature above 600 K to meet the optimal crystallization conditions for GST (typically, the growth rate of GST crystals reaches its highest point within the temperature range of 600-700 K) [38]. The insets in Fig. 5 provide a visual representation of the power-time profile of the pulse and the temperature distribution of the nanoscale antenna. Due to the presence of thermal boundary resistance, the heat loss of the PCM is minimized, thereby maintaining the temperature of the silver below its melting point. Therefore, the write/erase processes in this device are achievable and offer the advantages of fast response and low power consumption.

 

Fig. 5. Temperature curves during the amorphous and crystalline processes of the device. Insets (i) show the pulse details and (ii) the temperature profiles in amorphous and crystalline states.

Download Full Size | PDF

The aforementioned results show that the plasma device exhibits rapid switching capabilities and operates at low energy consumption. In comparison to other studies on photonic memory devices (Table 2), this work demonstrates significant improvements in terms of switching contrast and performance in changing the state of the PCM.

Table 2. Comparison of pulse parameter and energy for crystallization and amorphization.

View Table | View all tables in this article

3.3 Multi-level modulation and optical calculation test

Multi-level storage can significantly enhance the storage capacity and density of the memory. The crystallization level of GST can be adjusted by optical pulses with different duration and power. The crystallization level represents the mixture of amorphous and crystalline molecules at different proportions during the crystallization process [41]. The effective permittivity can be theoretically approximated using the effective medium theory:

The optical transmission at different crystallization levels is calculated, as shown in Fig. 6. It shows an almost linear relationship between optical transmission with the crystalline fraction in the cell. To achieve better linear transmission, we have chosen a range of crystallization levels from 20% to 65%. The fitted linear relationship based on the transmission (see inset in Fig. 6) can be calculated by using the formula as follows:

 

Fig. 6. The relationship between transmission and crystallization levels is controlled by multilevel adjustable switching.

Download Full Size | PDF

4. Device fabrication and tolerance

In this section, we outline the fabrication method and preparation tolerances for PCM devices that utilize the plasma-enhanced effect, as illustrated in Fig. 7. The manufacturing process begins by using electron beam lithography (EBL) to expose the required area on a Si substrate coated with a negative photoresist. Following this, the photoresist is developed, and inductively coupled plasma etching is performed on the silicon layer. Once the photoresist is removed, a layer of positive photoresist is coated on the chip surface. The elliptical metal rings are deposited by thermal evaporation on the metal pattern left after the photoresist is removed [43]. Subsequently, the photoresist is removed, resulting in a silicon waveguide device encapsulated by the metal rings. Finally, GST is deposited inside the metal ring and a layer of SiO₂ is coated on the plasma device to prevent oxidation [44].

 

Fig. 7. The schematic diagram of the fabrication process.

Download Full Size | PDF

Due to the non-ideal nature of the fabrication process, it is necessary to consider the influence of fabrication tolerances on the device. Figure 8(a) shows key device fabrication geometry including r_a and r_b of GST, Bragg width, and device positioning. Tolerances are simulated for these geometries. As evident from Fig. 8(b), we consider the impact of deviations in the size of the nanoscale antenna GST during the fabrication process, where the deviations are within 5 nm of our requirements. It is found that only when both parameters, r_a and r_b, deviate significantly, does it have a significant impact on the device. Furthermore, as shown in Fig. 8(c), the tolerances in both waveguide width (Δw < 5 nm) and the alignment of the waveguide device's center position (δ < 5 nm) are taken into consideration. The alignment has a controllable impact on contrast, while Δw significantly affects the F-P resonance wavelength. An observed maximum contrast difference of 8.5% is observed at Δw = ±5 nm. As depicted in Fig. 8(d), the resonant wavelengths under these conditions are 1531 nm and 1563 nm, respectively, corresponding to a maximum contrast of 62% and 69%. Therefore, the designed device exhibits high robustness.

 

Fig. 8. (a) Diagram of the three types of tolerance. (b) The contrast versus the Δr_a and Δr_b. (c) The optical contrast versus δ and Δw. (d) The resonance wavelength and corresponding maximum contrast occur when Δw is within the range of ±5 nm.

Download Full Size | PDF

5. Conclusion

In summary, we propose an innovative approach to combine an F-P resonator with a silver metal ring to amplify the interaction of GST and light, thereby significantly improving switching performance. Compared to conventional photonic memory devices, our device achieves 70.6% optical contrast in an ultra-compact size range, which is approximately 69 times higher than the contrast achieved through F-P cavity resonance alone. By employing the dual resonance approach, the device enables write and erase operations with remarkably low consumption for state switching. Leveraging improved optical contrast and precise heating phase transitions, our device enhances memory switching performance and storage density. Multilevel transmission can be obtained by modulating the fraction of amorphous and crystalline phases through adjusting the heating time. We validate the feasibility of our method through comprehensive process tolerance simulations, further enhancing the device's robustness. This work holds great potential for synthesizing photonic synapses with increased weight levels and higher resolution between adjacent weight levels, providing new solutions and potential applications in the field of optical computing.