Reconfigurable all-optical nonlinear activation function on silicon integrated platform

Abstract

This application relates to a reconfigurable all-optical nonlinear activation function on a silicon integrated platform. This application provides systems, apparatuses, and methods for an all-optical reconfigurable activation device that implements various activation functions using low input optical power. The disclosed apparatus and system include a directional coupler comprising a first phase-shifting mechanism and an interferometer coupled to the directional coupler. The interferometer includes at least one microring resonator and a second phase-shifting mechanism coupled thereto. The interferometer and directional coupler include a waveguide formed of a first material, while the microring resonator includes a waveguide formed of a second material and a third phase-shifting mechanism. The second material is provided as a low-loss material with a high Kerr effect and a large bandgap to generate various nonlinear activation functions. The first, second, and third phase-shifting mechanisms are configured to control bias within the disclosed system and apparatus to achieve the desired activation function.

Description

Background Technology

[0001] Driven by the growing interest in artificial intelligence (AI), the global market for artificial neural networks is projected to grow at a significant rate. Artificial neural networks (ANNs) and their learning algorithms possess the ability to learn from large datasets, potentially creating machines with low latency and high energy efficiency, exhibiting human-like decision-making capabilities. Compared to electronic systems, neuromorphic photonics demonstrates improved performance in multiplexing, energy dissipation, and crosstalk, which is advantageous for dense and high-bandwidth interconnects. Therefore, neuromorphic photonic systems have the potential to offer operating speeds several orders of magnitude faster and greater efficiency than neuromorphic electronic systems. ANNs are computational systems inspired by biological neural networks and consist of collections of connected nodes or neurons. Neurons include linearly weighted, summative, and nonlinear activations, with nonlinear activations being the building blocks of ANNs and enabling complex mappings between the inputs and outputs of the learning task. Several nonlinear functions, such as sigmoids, radial basis functions, rectified linear units (ReLUs), and quadratic functions, are widely used in ANNs for various machine learning tasks. Attached Figure Description

[0002] The present disclosure is described in detail with reference to the following accompanying drawings, which illustrate one or more various embodiments. The drawings are provided for illustrative purposes only and depict only typical or exemplary embodiments.

[0003] Figure 1 A model of an example nonlinear neuron including synapses, weighted addition, and nonlinear activation functions is shown according to the embodiments disclosed herein.

[0004] Figure 2 A schematic diagram of an all-optical nonlinear activation device according to an embodiment disclosed herein is depicted.

[0005] Figure 3A and 3B The example implementation is described as including Figure 2 An example resonant cavity in an all-optical nonlinear activated device.

[0006] Figure 3C Depicting Figure 3A and 3B Example mode conversion within an example resonant cavity.

[0007] Figures 4A-4C The example implementation is described as including Figure 2 Another example resonant cavity in an all-optical nonlinear activated device.

[0008] Figure 4D Depicting Figures 4A-4C Example mode conversion within an example resonant cavity.

[0009] Figure 5A-9C Graphical representations of various normalized nonlinear activation functions as functions of input optical power according to the embodiments disclosed herein are depicted.

[0010] Figure 10A and 10B An example phase shift mechanism including a metal oxide semiconductor capacitor (MOSCAP) according to an embodiment of the present disclosure is shown.

[0011] Figure 11 These are example computational components that can be used to implement various features of an all-optical nonlinear activation device according to the embodiments disclosed herein.

[0012] Figure 12 This is an example computer system that can be used to implement various features of the all-optical nonlinear activation device of this disclosure.

[0013] The accompanying drawings are not exhaustive and do not limit this disclosure to the precise form disclosed. Detailed Implementation

[0014] As mentioned above, ANNs and machine learning algorithms have the ability to learn from large datasets to create human-like machines. Neurons in an ANN consist of linearly weighted, summed, and nonlinearly activated inputs, where the nonlinear activations implement a complex mapping between the inputs and outputs used for learning. Examples of nonlinear activation functions include, but are not limited to, logistic functions, radial basis functions, linear rectified functions (such as ReLU, inverse ReLU, and leaky ReLU), and quadratic functions, each used for signal processing in different machine learning tasks. Various nonlinear activation functions are suitable for different tasks in neural networks and machine learning applications. For example, the ReLU function can be used to solve constrained nonlinear optimization problems and can be used in feedforward machine learning networks such as multilayer perceptrons and convolutional neural networks. Other examples include radial basis functions for multilayered support vector machines and quadratic functions for simulating high-order polynomial neural networks.

[0015] With the development of nonlinear optics, several all-optical methods have been proposed to realize activation functions. However, optical nonlinearity is relatively weak, so all optical activation devices typically require high threshold power and large light injection (e.g., inputting an optical signal into the device). Another technical drawback of traditional all-optical methods is that the activation devices are usually fixed after manufacturing and therefore cannot be configured to implement different activation functions. In other words, traditional activation devices are typically manufactured for a single activation function and cannot be switched (e.g., configured) for different activation functions.

[0016] Various methods have been employed to achieve nonlinear activation functions. These methods can generally be categorized into two types—photoelectric methods and all-optical methods. In photoelectric methods, nonlinear schemes integrated with photodetectors have been demonstrated based on silicon microring modulators, Mach-Zernd modulators (MZMs), electroabsorption modulators, or lasers. However, these photoelectric methods all have technical drawbacks; for example, all of them require efficient and fast photo-to-electric-to-optical conversion. These requirements almost invariably lead to increased system complexity and additional power consumption from electronic devices such as complementary metal-oxide-semiconductor (CMOS) field-effect transistors.

[0017] Therefore, all-optical methods have attracted increasing attention. Some implementations use microring resonators (MRRs) combined with phase change materials (PCMs) to alter the transmission-to-power relationship and achieve nonlinear activation functions. However, speed is limited due to constraints in the PCM. Based on the strong thermo-optical effect of germanium, germanium / silicon (Ge / Si) hybrid microring structures have been used to generate activation functions; however, the thermal process remains slow. All-optical schemes on silicon photonics platforms have been demonstrated, using Mach-Zenith interferometer (MZI) devices that utilize the free carrier dispersion (FCD) effect of silicon to provide nonlinearity. However, the free carriers induced by two-photon absorption (TPA) result in detrimental power and speed limitations for achieving certain activation functions. Modified versions of cavity-loaded MZI using the Si3N4 platform are still limited by the weak Kerr effect of Si3N4.

[0018] Therefore, the embodiments disclosed herein provide apparatus and methods for all-optically reconfigurable activation functions that overcome the drawbacks of the aforementioned methods. The embodiments disclosed herein utilize a heterogeneous silicon-on-insulator (SOI) platform, integrated with highly nonlinear and low-loss materials exhibiting a high Kerr effect and a large bandgap, to generate configurable all-optical nonlinear activation functions. Various embodiments provide resonant cavities using ultra-low-loss materials with a high Kerr effect and / or a large bandgap to enhance the nonlinearity of activation. In one example embodiment, aluminum gallium arsenide (AlGaAs) with a strong Kerr effect can be used to form the resonant cavity. In another example embodiment, tantalum pentoxide (Ta2O5), which also exhibits a strong Kerr effect, can be used. The Kerr effect (also known as the secondary electro-optic (QEO) effect) is a change in the refractive index of a material in response to an applied electric field (such as irradiation by a light source in the case of the optical Kerr effect).

[0019] According to the example embodiments disclosed herein, the all-optical reconfigurable activation device comprises a structure having a Mach-Zenith coupler (MZC) and a cavity-loaded Mach-Zenith interferometer (MZI). According to various embodiments, the resonant cavity is implemented as a microring resonator (MRR). Nonlinearity is provided by the resonant cavity, which alters the phase and amplitude of the transmission. The MZI converts the nonlinear phase into a nonlinear response. The phase of the MZI can be adjusted by a phase-shifting mechanism. The MZC acts as a tunable directional coupler, controlled by a phase-shifting mechanism optically coupled to a branch of the MZC. Various phase-shifting mechanisms provide programmability of the nonlinear function shape through a relative phase difference within the tuning structure; these can be any mechanism capable of inducing a phase shift in an optical signal propagating through a corresponding waveguide. For example, the resonant wavelength of the resonant structure can be tuned / detuned and the coupling coefficients can be tuned / detuned to switch between various activation functions (e.g., reconfiguration), such as logic functions, radial basis functions, ReLU (e.g., ReLU, inverse ReLU, and leaky ReLU), and quadratic functions for different task applications.

[0020] In an example implementation, the cavity-loaded MZI includes a first waveguide formed of a first material coupled to a second waveguide formed of a second material. The second waveguide includes an input and an output, wherein the input of the second waveguide is coupled to the output of the first waveguide and the output of the second waveguide is coupled to the input of the first waveguide. In some implementations, to optimize mode switching between the first and second waveguides, an optical signal propagating in the first waveguide is briefly coupled to the second waveguide via a first inverted cone pair (e.g., each of the first and second waveguides tapers in opposite directions relative to each other). Similarly, an optical signal propagating in the second waveguide is briefly coupled to the first waveguide via a second inverted cone pair. The first and second waveguides may form branches of the MZI. The resonant cavity formed of the second material is briefly coupled to the second waveguide.

[0021] According to the embodiments disclosed herein, the second material is provided as a low-loss material with a high Kerr effect and a large bandgap to generate a configurable all-optical nonlinear activation function. For example, the second material may be a material with a 2.6 x 10⁻⁶ bandgap. -17 m 2 AlGaAs with a nonlinear refractive index of / W and a band gap of 1.92 eV. In another example, the second material could be AlGaAs with a refractive index of 7.2 x 10⁻⁶ W and a band gap of 1.92 eV. - 19 m 2 Ta₂O₅ has a nonlinear refractive index of / W and a band gap of 3.8 eV. The first material, according to various embodiments, can be silicon or another group VI material. It should be noted that the nonlinear refractive index of silicon is 4.5 x 10⁻⁶. -18 m 2 / W and a band gap of 1.12eV.

[0022] The high Kerr effect of the second material allows for significantly lower optical injection power than other all-optical platforms, such as those that rely on silicon-based FCDs to provide nonlinear activation. The first material also includes thermal effects due to negligible high power intensity. Furthermore, the absence of two-photon absorption (TPA) and induced free carrier dispersion (FCD) due to the large bandgap eliminates the velocity limitation from carrier lifetime. For the first material, the FCD conflicts with the Kerr effect due to the presence of free carriers generated by TPA. For example, in the case of silicon, the FCD and Kerr effect in silicon would drive the resonant frequency in opposite directions. However, the FCD generally dominates the Kerr effect, resulting in a reduction in overall nonlinearity in the case of silicon. The second material according to the embodiments disclosed herein can be used to eliminate this conflict.

[0023] It should be noted that the terms “optimization,” “optimal,” etc., as used herein, can be used to mean achieving the most efficient, perfect, or improved performance possible. However, as a person skilled in the art who reads this document will recognize, perfection cannot always be achieved. Therefore, these terms can also include achieving the best possible or most efficient performance under given conditions, or achieving performance that is better than that achievable using other settings or parameters.

[0024] As used herein, “approximately” and “generally” refer to permissible variations in the properties of the embodiments disclosed herein. The embodiments disclosed herein may have certain properties, features, and / or characteristics, including some acceptable variations that do not significantly affect the functionality of the disclosed embodiments.

[0025] Figure 1 An example neuron of an ANN is shown, where the inputs entering the neuron (e.g., X1 to X) n This is a linear combination (weighted addition) of the outputs of other neurons. This neuron assigns weights (e.g., W1 to W2) to the outputs of other neurons. n The input signal is applied and weighted over time (e.g., summed) to produce a nonlinear response Y, which is represented by an activation function. The neuron's output is then broadcast to successive neurons in the ANN. Note that the connections between neurons can be weighted using positive and negative values, respectively, represented as excitatory and inhibitory synapses. The synaptic interconnection network of neurons can be represented as weight values ​​(W). ij It can be a matrix of real numbers. Furthermore, the encoding scheme maps real-valued weights and represents them as spike signals.

[0026] Figure 2A schematic diagram of an all-optical nonlinear activation device 100 according to an embodiment disclosed herein is depicted. The activation device 100 includes an input terminal for receiving an input optical signal and an output terminal for transmitting an output optical signal. The activation device 100 includes a Mach-Zernd coupler (MZC) 106 that receives the input optical signal and is optically coupled to the output optical signal via a Mach-Zernd interferometer (MZI) 110. The MZC 106 includes branches 118 and 120, each of which can be implemented as a waveguide guiding the propagation of light (e.g., an optical signal such as a laser mode). The MZC 106 includes a first phase shift mechanism 108 in one of the branches 118 and 120. Figure 2 In the illustrative example shown, the first phase shift mechanism 108 is provided along branch 118; however, the first phase shift mechanism 108 may be provided along either branch 118 or branch 120. MZI 110 includes branches 122 and 124, each of which may be implemented as a waveguide for guiding the propagation of light (e.g., an optical signal such as a laser mode). MZI 110 also includes a second phase shift mechanism 112 and at least one MRR 114, wherein each of the second phase shift mechanism 112 and at least one MRR 114 is optically coupled to one of branches 122 and 124 of MZI 110. The second phase shift mechanism 112 and MRR 114 may be optically coupled along the same (or common) branch of MZI 110. In the illustrative example, the second phase shift mechanism 112 and MRR 114 are provided along branch 122; however, the second phase shift mechanism 112 may be provided along either branch 122 or branch 124. Additional MRRs may be included in another branch, or multiple MRRs may be included in one branch, depending on the implementation. In some implementations, MRR 114 may include a third phase shift mechanism 116.

[0027] Phase-shifting mechanisms 108, 112, and 116 are configured to change the phase of the optical signal propagating therein. Phase-shifting mechanisms 108, 112, and 116 can be provided as any mechanism capable of inducing a phase shift in light propagating through the respective waveguide (specific examples of phase-shifting mechanisms are provided in more detail below). Figure 2In the illustrative example, a first phase shift mechanism 108 can be controlled to tune the relative phase difference between branches 118 and 120 of MZC 106 by inducing a phase shift in branch 118. A second phase shift mechanism 112 can be controlled to tune the relative phase difference between branches 122 and 124 of MZI 110 by inducing a phase shift in branch 122. A third phase shift mechanism 116 can be controlled to tune the resonant frequency of MRR 114 by inducing a phase shift in the resonant cavity (e.g., waveguide) of MRR 114. By tuning the relative phase difference within the activation device 100, phase shift mechanisms 108, 112, and / or 116 serve as tunable elements configured to effectively change the bias of the activation device 100, enabling the device 100 to be programmed to implement desired activation functions for different applications (as described below). Figure 5A-9C (As described). For example, one or more of the tuning phase shift mechanisms 108, 112, and / or 116 provide switching between different activation functions. Therefore, controlled tuning of the phase shift mechanisms 108, 112, and / or 116 provides the configuration of the activated device to a desired activation function, which can be changed at a later time by controlling the phase shift mechanisms 108, 112, and / or 116. Since the tunable element can be controlled by an automatic control system (e.g., implemented as... Figure 12 The computer system 1200 precisely controls the device, so the activation device 100 can perform these functions with high precision.

[0028] According to various embodiments, MZC 106 serves as a tunable directional coupler based on phase shift mechanism 108. For example, tuning the relative phase difference between branches 118 and 120 of MZC 106 via phase shift mechanism 108 provides tuning of the splitting ratio (r) of the optical power supplied to each branch of MZI 110. That is, the ratio of the optical power supplied to branch 122 of MZI 110 to the optical power supplied to branch 122 can be controlled by tuning the phase difference between the optical signal in branch 118 and the optical signal in branch 120 via phase shift mechanism 108. For example, when the splitting ratio (r) is 1, all the optical power from MZC 106 (e.g., the optical power in both branches 118 and 120) is supplied to branch 124. A splitting ratio of 1 can be achieved by tuning the relative phase difference between branches 118 and 120 of MZC 106 to 270° (e.g., 3π / 2 radians). Furthermore, when the splitting ratio (r) is zero, all optical power from MZC 106 (e.g., the optical power from both branches 118 and 120) is supplied to branch 122. A splitting ratio of 0 can be achieved by tuning the relative phase difference between branches 118 and 120 of MZC 106 to 90° (e.g., π / 2 radians). Therefore, the phase shift mechanism 108 can be controlled to tune the relative amplitude of the optical power supplied to branches 122 and 124 of MZI 110.

[0029] In the MRR 114, when the optical signal approaches its resonant frequency, it undergoes a power-dependent nonlinear phase shift that changes rapidly with its optical power. Furthermore, the MRR 114 can increase the effective interaction length and instantaneous optical power through coherent power accumulation, thereby reducing the required optical power supply. The phase shift mechanism 116 can be controlled to tune the resonant frequency. The MZI 110 is used to convert the phase change from the MRR 114 into an intensity change with a large extinction ratio. With a sufficiently large phase difference, interference between the optical signals propagating in the two branches of the MZI 110 can switch from constructive to destructive, resulting in self-switching. Therefore, as... Figure 2 As shown, loading at least one branch of MZI 110 with MRR 114 provides configurability of the all-optical nonlinear activation device 100 between different activation functions. Additional MRRs can be provided to support and / or assist this switching effect. Therefore, while multiple MRRs are possible, at least one branch is loaded with at least one MRR to achieve the desired effect.

[0030] To maximize the switching effect, low-power signals can be shut off through destructive interference. Ideal destructive interference requires signals traveling in both branches of the MZI 110 to have equal amplitude and a precise π-phase difference. As described above, the MZC 106 preceding the MRR-assisted MZI 110 serves as a tunable directional coupler, configured to tune the amplitude of the optical power supplied to each branch of the MZI 110 via a phase-shifting mechanism 108. The bias of the MZC 106 (via phase-shifting mechanism 108) can be adjusted to balance the amplitudes at both branches of the MZI 110, while the bias of the MZI 110 can be individually tuned to introduce a π-phase difference to achieve destructive interference. The bias on the MRR 114 can also be adjusted to ensure that the activated device 100 operates approximately at the resonant wavelength to achieve optimal sensitivity (e.g., the highest possible sensitivity).

[0031] The dynamic response of the activation device 100 can be simulated based on the rate equation and coupling mode theory described below. The amplitude of the optical signal in the MRR 114 (e.g., calculated as...) The magnitude of the change over time) and the free carrier (FC) density N (e.g., calculated as...) The change in FC density over time can be used to simplify the dynamic equation:

[0032]

[0033]

[0034] Where j represents an imaginary value; λ represents the wavelength of the input optical signal (e.g., 1310 nm in some examples); Δλ represents the detuning of the wavelength of the input optical signal from the resonant frequency of the MRR 114; c is the speed of light; γ L ,μ,P in and τ fc η represents the linear decay rate, the energy coupling coefficient of MRR 114 (e.g., 33000 in the case of AlGaAs), the input power, and the free carrier lifetime (e.g., 0.5 ns in some examples); k η fc ξ and ξ are related to the Kerr effect coefficient, the free carrier dispersion (FCD) effect coefficient, and the two-photon absorption (TPA) coefficient, respectively; a represents the half-frame amplitude in MRR 114; |a| 2 Let N represent the power in MRR114; N represents the free carrier density in MRR114. In Equation 1, Representing linear effects, and jη k |a| 2 a-jη fc Na represents the nonlinear effect.

[0035] Linear decay rate (γ) L ), input power (P) in ), Kerr effect coefficient (η) k ), FCD effect coefficient (η) fc The free carrier generation coefficient (ξ) and the free carrier generation coefficient (ξ) are determined as follows:

[0036]

[0037]

[0038]

[0039]

[0040]

[0041] Where v g This represents the group velocity (e.g., approximately 7.9 x 10^6 in the case of AlGaAs). 7 m / s); α0 represents the linear loss coefficient; T rt This represents the round-trip time in the micro-ring (e.g., 9.56 ps in the case of AlGaAs); n c Indicates the core refractive index; n r The valid index of the mode is indicated; n2 represents the nonlinear refractive index; α2 represents the two-photon absorption (TPA) coefficient (e.g., this coefficient is approximately 0 in the case of low-loss and large-bandgap materials); A eff Indicates the area of ​​the effect mode (e.g., 0.15 μm). 2 ); k0 represents the wave number; σ r Represents the free carrier refractive volume (e.g., -5.3 x 10⁻⁶). 27 m 3 );σ a This represents the free carrier absorption cross section (1.5 x 10⁻⁶). 21 m 2 ).

[0042] According to Equations 1 and 2, intensity-related nonlinearity originates from the Kerr effect, FCD, and / or TPA. For example, as shown in Equation 1, nonlinear effects include those dependent on the Kerr effect coefficient (η). k The first part and the effect coefficient (η) depend on the FCD effect coefficient. fc The second part of the free carrier generation coefficient (ξ) and FC density (N) depends on the free carrier generation coefficient (ξ) as shown in Equation 1. Kerr effect coefficient (η) kThe resonant frequency of MRR 114 is proportional to the nonlinear refractive index (n2), and the TPA coefficient (ξ) is proportional to the two-photon absorption coefficient (α2). When the Kerr effect dominates (because n2 > 0), the resonant frequency of MRR 114 is redshifted (frequency and photon energy decrease, while wavelength increases), while the FCD effect is in the opposite direction because the FC volume is negative.

[0043] Table 1 below provides a list of various materials that can be implemented in the activation device 100. Table 1 provides the core refractive index (n) of each material. c ); nonlinear refractive index (n2), linear loss coefficient (α0), and the ratio of nonlinear refractive index (n2) to linear loss coefficient (α0).

[0044] Table 1

[0045] <![CDATA[ Material ]]> <![CDATA[ n c ]]> <![CDATA[ Band gap (eV) ]]> <![CDATA[ n2(m 2 / W) ]]> <![CDATA[ α0(dB / m) ]]> <![CDATA[ n2 / α0 ]]> Si 3.45 1.12 <![CDATA[4.5×10 -18 ]]> 200 <![CDATA[2.25×10 -20 ]]> InGaP 3.1 1.9 <![CDATA[4×10 -18 ]]> 1000 <![CDATA[4×10 -21 ]]> AlN 2.12 6 <![CDATA[2.3×10 -19 ]]> 50 <![CDATA[4.6×10 -21 ]]> <![CDATA[Si3N4]]> 2 5.1 <![CDATA[2.4×10 -19 ]]> 8 <![CDATA[3×10 -20 ]]> InGaAsP 3.58 1.2 <![CDATA[1×10 -17 ]]> 300 <![CDATA[3.33×10 -20 ]]> GaAs 3.4 1.42 <![CDATA[2×10 -17 ]]> 50 <![CDATA[4×10 -20 ]]> GaP 3.05 2.24 <![CDATA[6×10 -18 ]]> 140 <![CDATA[4.29×10 -20 ]]> <![CDATA[Ta2O5]]> 2.06 3.8 <![CDATA[7.2×10 -19 ]]> 8 <![CDATA[9×10 -20 ]]> AlGaAs 3.28 1.92 <![CDATA[2.6×10 -17 ]]> 60 <![CDATA[4.33×10 -19 ]]>

[0046] As mentioned above, conventional all-optical activated devices rely on the FCD effect of silicon to provide nonlinearity using an MRR formed from silicon waveguides. However, as can be derived from Equations 1-7, this nonlinear effect is weak and requires very high input optical power to achieve certain activation functions. Furthermore, due to the presence of free carriers generated by the TPA, the FCD effect conflicts with the Kerr effect (e.g., the FCD and Kerr effects drive the resonant frequency in opposite directions), and the FCD generally dominates over the Kerr effect. MRRs formed from Si3N4 waveguides are still limited by the weak Kerr effect of Si3N4.

[0047] Therefore, the activation device 100 according to the embodiments disclosed herein includes MZC 106 and MZI 110 having branches including waveguides formed of a first material, and MRR 114 having waveguides formed of a second material dissimilar to the first material. The first material may be silicon or another group IV material (e.g., germanium, silicon carbide, silicon germanium, etc.). The second material is a low-loss material with high Kerr effect and low FCD and / or TPA effect. The second material is selected from materials having a band gap greater than twice the operating wavelength (e.g., 1.3 μm to 0.95 eV). As a result of this characteristic, the TPA effect is negligible. Materials with negligible TPA also typically have a low linear loss coefficient (α0). However, due to manufacturing, processing, or other integration issues entering the SOI platform, the linear loss coefficients of each material change as shown in Table 1. Materials with narrow band gaps (e.g., such as silicon) will result in free carrier generation. Therefore, for these narrow bandgap materials, the thermal problems caused by charge carriers at high optical power cannot be ignored, and thus the waveguide cannot support high-power optical signals. The free carrier effect is a slow process (< GHz) and limits processing speed due to the lifetime of free carriers. The Kerr effect is a fast process (easily reaching over 10 GHz) and can improve data speed. As mentioned above, the FCD effect is the opposite of the Kerr effect, but the contrast can be avoided if there are no free carriers. Therefore, the embodiments disclosed herein provide a second material selected from materials having a bandgap of at least greater than 1.12 eV and more preferably greater than 1.9 eV. Therefore, as used herein, wide bandgap or large bandgap means a bandgap of at least greater than 1.12 eV.

[0048] According to various embodiments, a second material can be selected from materials having a desired bandgap, as a material having at least one of the following: a high nonlinear refractive index (n2) and a low linear loss coefficient (α0). For example, the embodiments disclosed herein provide a second material having at least one of the following: (i) a nonlinear refractive index (n2) higher than that of a silicon waveguide and (ii) a linear loss coefficient (α0) lower than that of a silicon waveguide. As used herein, a high nonlinear refractive index (n2) implies nonlinearity due to a strong Kerr effect and a low required input optical power. For example, a high nonlinear refractive index (n2) (e.g., a strong Kerr effect) refers to having a value greater than 2.4 x 10⁻⁶. -19 m 2 Materials with a nonlinear refractive index (n²) of / W, and a very high nonlinear refractive index (n²) (e.g., a very strong Kerr effect) refers to materials with a nonlinear refractive index greater than 4.5 x 10⁻⁶. -18 m 2A material with a nonlinear refractive index (n²) of / W. As used herein, a low linear loss factor (α₀) means a high quality factor for the MRR 114 and enhanced optical power in the MRR 114. For example, a low linear loss factor (α₀) refers to a linear loss factor (α₀) below 200 dB / m, such that the quality factor of the MRR is higher than that of a silicon-based MRR and the optical power in the MRR 114 is enhanced relative to a silicon-based MRR. Furthermore, the TPA of the low-loss material is negligible, and therefore thermally induced offset can be ignored. Thus, the embodiments disclosed herein can support high input power optical signals. The FCD from the TPA, which limits the processing speed due to the lifetime of free carriers, can also be ignored, and the operating speed can be improved.

[0049] In some embodiments, the second material can be selected based on both the nonlinear refractive index (n2) and the linear loss coefficient (α0) to provide a nonlinear effect dependent on the Kerr effect, while the TPA and FCD effects are negligible. For example, as mentioned above, the TPA effect dominates and controls the nonlinear effect, while the FCD effect causes velocity limitation. Therefore, some disclosed embodiments use a second material with a wide bandgap (e.g., exceeding 1.12 eV, more preferably exceeding 1.9 eV) to avoid two-photon absorption at 1310 nm. Furthermore, the second material has a ratio n2 / α0 greater than that of silicon. More specifically, this ratio n2 / α0 can be greater than 5 x 10⁻⁶. -20 In some embodiments, the ratio of the second material, n2 / α0, is at least 9 × 10⁻⁶. -20 For example, in the case of Ta2O5 and AlGaAs. The following combines... Figures 3A-3C Non-limiting examples of activation devices 100 implemented using MRR 114, including waveguides formed of Ta2O5 and AlGaAs, are provided in 4A-4D, respectively.

[0050] Return to Figure 2 An optical signal propagating along branch 122 of MZI 110 can be briefly coupled to MRR 114 based on the coupling coefficient between the waveguide of branch 122 and the waveguide of MRR 114. The coupling efficiency can be affected by the difference in waveguide materials between branch 122 and MRR 114. As described above, each branch of MZC 106 and MZI 110 comprises a waveguide formed of a first material (e.g., silicon or other Group IV materials), while MRR 114 is formed of a closed-loop waveguide formed of a second material, which is a low-loss material with a high Kerr effect and a large bandgap. However, due to the difference in material properties, a large phase difference can exist between the waveguides of branch 122 and MRR 114.

[0051] To achieve a sufficient coupling coefficient for critical coupling of the MRR, a second waveguide 128 may be included in branch 122, configured to facilitate optimal mode transitions from branch 122 to MRR 114 and vice versa. The sufficient coupling coefficient for critical coupling conditions, as used herein, refers to a coupling coefficient equal to the loss per round in the microring. According to some embodiments disclosed herein, the sufficient coupling coefficient may be less than 1%, depending on the material loss and length of MRR 114. For example, the second waveguide 128 may be formed of a second material and coupled to a first waveguide formed of a first material at a location adjacent to MRR 114 along branch 122. In some embodiments (as will be described in more detail below), to optimize mode transitions between the first and second waveguides of branch 122, an inverted tapered pair (e.g., each of the first and second waveguides tapers in opposite directions relative to each other) may be provided at the input / output of each waveguide. The MRR 114, formed of the second material, is briefly coupled to the second waveguide 128. Therefore, by optimizing the mode conversion between the first and second waveguides of branch 122, a sufficient coupling coefficient between MRR 114 and branch 122 can be achieved.

[0052] Figure 3A and 3B An example resonant cavity included in an all-optical nonlinear activation device 100 according to an exemplary embodiment is depicted. Specifically, Figure 3A and 3B Depicting Figure 2 An enlarged perspective view of region 130, which includes a resonant cavity (e.g., MRR 314) coupled to branch 122 of the activation device 100. Figure 3A In the illustrative example, branch 122 of MZI 110 includes a first waveguide 302 briefly coupled to a second waveguide 328. MRR 314 is briefly coupled to the second waveguide 328. The optical signal propagating in MZI 110 is briefly coupled from the first waveguide 302 to the second waveguide 328, from the second waveguide 328 to MRR 314, back to the second waveguide 328, and then from the second waveguide 328 to the first waveguide 302. Figure 3B The image depicts the view observed from the side opposite MRR 114 along branch 122. Figure 3A Side view.

[0053] like Figure 3A and 3B As shown, the activation device 100 can have various widths and thicknesses. As used herein, it can be along... Figure 3A and 3B The "length" is defined by the X-axis, which can be used along... Figure 3A and 3B The "width" is defined by the Y-axis, and can be defined along... Figure 3A and 3B The Z-axis is used to define "height" or "thickness".

[0054] exist Figure 3A and 3B In the example shown, MRR 314 and the second waveguide 328 are formed of TA2O5 as the second material. In this illustrative example, the second waveguide 328 and MRR 314 are formed on an SOI substrate 320, wherein the SOI substrate 320 has a buried oxide (BOX) layer 324 formed thereon. To form the first waveguide 302, a first material (e.g., silicon or other Group IV material) is deposited on the patterned and etched BOX layer 324. According to various embodiments, the first material is silicon. Although in Figure 3A and 3B Only region 130 is shown, but MZC 106 and MZI 110 are formed using a similar process. For example, the entire activation device 100 can be formed by providing an SOI substrate with a BOX layer formed thereon, followed by depositing a first material to form MZC 106 and MZI 110 (e.g., waveguides constituting MZC 106 and MZI 110).

[0055] After patterning and etching of the waveguide structure of MZI 110, hydrogen silsesquioxane (HSQ) planarization is performed. For example, HSQ planarization is achieved by spin-coating an HSQ photoresist layer to activate the apparatus 100, followed by rapid thermal annealing to convert the HSQ photoresist layer into a silicon dioxide (SiO2) layer 326. The silicon dioxide layer 326 can be patterned and locally etched to a desired distance (e.g., height) from the first waveguide 302. A tantala film can be deposited onto the etched layer 326 via sputtering. The tantala film, according to various embodiments, can be a TA2O5 film or a coating. The tantala film can undergo thermal annealing, which relieves stress and reduces optical loss within the second waveguide 328. Optionally, the first waveguide 302 and the second waveguide 328 can be positioned along a stacking direction (e.g., in...). Figure 3B The desired distance (in the context of the Z-axis) is used to achieve the desired coupling coefficient between the uppermost surface of the first waveguide 302 and the lowermost surface of the second waveguide 328. For example, the distance between them can be chosen to optimize the coupling coefficient.

[0056] In the illustrative example, the activation device 100 includes a fully etched 500 nm wide silicon waveguide (e.g., MZC 106 and MZI 110 including a first waveguide 302) and a 1 μm wide TA2O5 waveguide (e.g., MRR 314 and a second waveguide 328) with a material thickness of 1 μm on a passive SOI platform. The BOX layer 324 may be a 3 μm thick oxide passivation layer. The MRR 314 may have a radius of 20 μm and a power coupling factor of approximately 0.4%, with a gap of approximately 340 nm between the second waveguide 328 and the MRR 314, producing a Q factor of approximately 160,000. The distance in the Z-axis direction between the first waveguide 302 and the second waveguide 328 is approximately 100 nm, which can provide a coupling factor exceeding 97%. Although in Figure 3A and 3B It is not shown in the figure, but MRR 314 can be coupled to phase shift mechanism 116.

[0057] Figure 3A Images of fundamental transverse (TE) modes 330 and 332 of the optical signal propagating in the first waveguide 302 and the second waveguide 328, respectively, are depicted. Fundamental TE mode 330 corresponds to the optical signal propagating in the first waveguide 302 at line 338, and fundamental TE mode 332 corresponds to the optical signal propagating in waveguide 328 at line 340. It should be noted that the effective refractive indices of silicon and TA2O5 are 2.78 and 1.9, respectively. As shown in fundamental TE modes 330 and 332, a large phase mismatch occurs in the optical signal propagating in each corresponding material, making it difficult to achieve a sufficient coupling coefficient for critical coupling. Therefore, in Figure 3A In the example embodiment shown, in order to optimize the mode conversion between the first waveguide 302 and the second waveguide 328, a first inverted cone pair 336 is provided to optimize the transient coupling of the optical signal propagating in the first waveguide 302 to the second waveguide 328, and a second inverted cone pair 318 is provided to optimize the transient coupling of the optical signal propagating in the second waveguide 328 to the first waveguide 302.

[0058] For example, the first waveguide 302 may have a width W1 at the input junction 312 of the second waveguide 328 (e.g., the optical signal closest to being coupled into the second waveguide 328) and a width W2 at the output junction 316 of the second waveguide 328 (e.g., the optical signal closest to being coupled out of the second waveguide 328). In some embodiments, widths W1 and W2 may be the same or substantially similar in width. The second waveguide 328 may have a width W3 between the output junction 306 and the input junction 310 of the first waveguide 302. The output junction 306 may have a width W4, the input junction 310 may have a width W5, the output junction 316 may have a width W6, and the input junction 312 may have a width W7. In some embodiments, widths W1 and W2 may be the same or substantially similar in width, widths W4 and W5 may be the same or substantially similar in width, and widths W6 and W7 may be the same or substantially similar in width. In the above example implementation, W1 and W2 can be 500nm; W3 can be 1μm; and W4 to W7 can be 200nm.

[0059] Each inverted cone pair 336 and 318 can be implemented as a corresponding mode converter (e.g., mode converter 336 and mode converter 318, respectively). The first inverted cone pair 336 includes a cone 304 that narrows the width W1 to a width W4 at the output junction 306 of the first waveguide 302 and a cone 342 that widens the width W6 at the input junction 312 to a width W3. The second inverted cone pair 318 includes a cone 308 that widens the width W5 at the input junction 310 of the first waveguide 302 to a width W2 and a cone 344 that narrows the width W3 to a width W7 at the output junction 316 of the second waveguide 328. (The text repeats itself here.) Figure 3A As shown, portions of the first waveguide 302 and the second waveguide 328 overlap in the Z-axis direction within a first conical region 356 having a length L1 corresponding to the first inverted conical pair 336, and portions of the first waveguide 302 and the second waveguide 328 overlap in the Z-axis direction within a second conical region 358 having a length L2 corresponding to the second inverted conical pair 318. Figure 3A As shown, portions of the first waveguide 302 and the second waveguide 328 overlap in the Z-axis direction within a first conical region 356 having a length L1 corresponding to the first inverted cone pair 336, and portions of the first waveguide 302 and the second waveguide 328 overlap in the Z-axis direction within a second conical region 358 having a length L2 corresponding to the second inverted cone pair 318. In some embodiments, lengths L1 and L2 may be the same or substantially similar in width. In an example embodiment, L1 and L2 may be 60 μm.

[0060] The first inverted cone pair 336 facilitates the coupling of the fundamental TE mode between the first waveguide 302 and the second waveguide 328, while the second inverted cone pair 318 facilitates the coupling of the fundamental TE mode between the second waveguide 328 and the first waveguide 302. Optical coupling at each corresponding inverted cone pair 336 and 318 can be a result of the cones included therein. That is, the cone of the first inverted cone pair 336 can push the fundamental TE mode upwards into the second waveguide 328, allowing it to be effectively coupled into the MRR 314, while the cone of the second inverted cone pair 318 can push the fundamental TE mode downstream in the first waveguide 302 and the MZI110. The first inverted cone pair 336 and the second inverted cone pair 318 can be designed such that they each satisfy certain performance characteristics. For example, the first inverted cone pair 336 and / or the second inverted cone pair 318 can be designed to be thermally insulating. As another example, the first inverted cone pair 336 and / or the second inverted cone pair 318 can be designed such that they are respectively coupled to a basic TE mode with very low loss and low back reflection. Mode converters implemented as having inverted cone pairs that are too short in length may suffer from high passive losses.

[0061] Figure 3C An example of mode conversion during the coupling of an optical signal from a first waveguide 302 to a second waveguide 328 within a first conical region 356 viewed along branch 122 is depicted. For example, Figure 3C The light intensities of the basic TE modes 330 and 332 along the X-axis direction are depicted. For the first inverted cone pair 336 with a cone length L1 of 60 μm, the coupling efficiency according to the embodiment disclosed herein can exceed 97%. Because the second waveguide 328 is wider, the modes in the second waveguide 328 appear less concentrated compared to the modes in the first waveguide 302.

[0062] Compared to the bandgap of silicon (e.g., 1.12 eV), Ta₂O₅ has a larger bandgap (e.g., 3.8 eV), therefore, according to Figures 3A-3C The second material in the example implementation includes Ta2O5. Therefore, the TPA effect and the FCD effect caused by TPA are negligible, for example, on optical signals with a wavelength of 1310 nm and do not provide nonlinearity. Furthermore, optical loss is low due to the absence of free carrier absorption. For example, the linear loss coefficient is as low as 3 dB / m for a straight waveguide formed from Ta2O5, and 8 dB / m for an MRR with a radius of 20 μm formed from Ta2O5. The nonlinear refractive index of Ta2O5 is three times higher than that of another low-loss material, Si3N4 (e.g., n2 of Si3N4 is approximately 2.4 x 10⁻⁶). -19 m 2 / W, in comparison, the n2 of Ta2O5 is approximately 7.2 x 10⁻⁶. -19 m2 / W4), which allows for a reduction in the input optical power to the activation device 100 to achieve a nonlinear effect.

[0063] Figures 4A-4C Another example resonant cavity, included in an all-optical nonlinear activation device 100 according to an exemplary embodiment, is depicted. Specifically, Figures 4A-4C It describes the resonant cavity. Figure 2 An enlarged perspective view of region 130, such as example MRR 414, coupled to branch 122 of activation device 100. Figure 4A In the illustrative example shown, branch 122 of MZI 110 includes a first waveguide 402 briefly coupled to a ribbed waveguide 464 comprising a second waveguide 428 and a second waveguide plate 454. MRR 414 includes waveguide 460 and waveguide plate 462, which together form a ribbed waveguide briefly coupled to the second waveguide 428. An optical signal propagating in MZI 110 is briefly coupled from the first waveguide 402 to the second waveguide 428, from the second waveguide 428 to the MRR 414, returns to the second waveguide 428, and then is coupled from the second waveguide 428 to the first waveguide 402. Figure 4B The image depicts the view observed from the side opposite MRR 414 along branch 122. Figure 4A The side view, and Figure 4C The image depicts the view along line 440 from the side opposite to MRR 414. Figure 4A Cross-sectional side view.

[0064] like Figure 4A and 4C As shown, the activation device 100 can have various widths and thicknesses. As used herein, it can be along... Figure 4A and 4C The "length" is defined by the X-axis, which can be used along... Figure 4A and 4B The "width" is defined by the Y-axis, and can be defined along... Figure 4A and 4C The Z-axis is used to define "height" or "thickness".

[0065] exist Figures 4A-4CIn the example shown, MRR 414 and ribbed waveguide 464 are formed from a heterogeneous III-V material (e.g., AlGaAs in this example) as the second material. In this illustrative example, the ribbed waveguide 464 and MRR 414 are heterogeneously formed on an SOI substrate 420, on which a buried oxide (BOX) layer 424 is formed. For example, the epitaxial second material can be grown on a III-V substrate (e.g., the GaAs substrate in this example) and bonded to the SOI platform via a bonding layer 452. In some example embodiments, an Al2O3 or HfO2 dielectric layer can be used as the bonding layer 452. To form the first waveguide 402, a first material (e.g., silicon or a Group IV material) is deposited on the patterned and etched BOX layer 424. In various embodiments, the first material is silicon. Although in Figures 4A-4C Only region 130 is shown, but MZC 106 and MZI 110 are formed using a similar process. For example, the entire activation device 100 can be formed by providing an SOI substrate with a BOX layer formed thereon, and then depositing a first material to form MZC 106 and MZI 110 (e.g., waveguides constituting MZC 106 and MZI 110).

[0066] The bonding layer 452 may have a certain thickness to provide the desired distance between the first waveguide 402 and the ribbed waveguide 464. This can be selected along the stacking direction (e.g., in...). Figure 4B and 4C The desired distance between the first waveguide 402 and the ribbed waveguide 464 (in the context of the Z-axis) is determined to achieve a desired coupling coefficient between the uppermost surface of the first waveguide 402 and the bottommost surface of the ribbed waveguide 464. For example, the distance between them can be chosen to optimize the coupling coefficient.

[0067] In various embodiments, a plurality of support structures 446 and 448 configured to support the second waveguide 428 are provided. For example, a first support structure 446 and a second support structure 448 are formed on layer 424. A bonding layer 452 is formed on top of the first support structure 446 and the second support structure 448, thereby forming an air trench 450. The air trench 450 provides mode limitation for optical signals within the respective waveguides 428 and / or 402, i.e., the air trench 450 restricts the mode of optical signal propagation within the first waveguide 402 and / or the second waveguide 428.

[0068] In the illustrative example, the activation device 100 includes a fully etched 500 nm wide silicon waveguide (e.g., MZC 106 and MZI 110 including a first waveguide 402) and a 600 nm wide AlGaAs waveguide (e.g., waveguides 428 and 460) with a thickness of 190 nm and ribbed waveguides (e.g., ribbed waveguides 464 and MRR 414) on a passive SOI platform with a material thickness of 220 nm. Waveguides 428 and 460 may be 150 nm thick, and the corresponding plates 462 and 454 may be 40 nm thick. Furthermore, plates 454 and 462 may extend 2 μm from either side of the corresponding waveguides 428 and 460, such that each rib structure has a width of 4.6 μm. The BOX layer 424 may be a 3 μm thick oxide passivation layer, the bonding layer 452 may be 15 nm thick, and the III-V group may be 600 μm thick. MRR 414 can have a radius of 12 μm and a coupling factor of approximately 0.8%, with a gap of approximately 270 nm between the second waveguide 428 and MRR 414, producing a Q factor of approximately 70,000. The distance in the Z-axis direction between the upper surface of the first waveguide 402 and the lower surface of the ribbed waveguide 464 is approximately 15 nm (e.g., the distance to the bonding layer), which can provide a coupling factor of approximately 98%. Although not in Figures 4A-4C As shown, however, MRR 414 can be coupled to phase shift mechanism 116.

[0069] Figure 4A Images include fundamental lateral (TE) modes 430 and 432 of the optical signals propagating in the first waveguide 402 and the second waveguide 428, respectively. Fundamental TE mode 430 corresponds to the optical signal propagating in the first waveguide 402 at line 438, and fundamental TE mode 432 corresponds to the optical signal propagating in waveguide 428 at line 440. As described above, modes 430 and 432 are confined to their respective waveguides via air trenches 450. It should be noted that the effective refractive indices of silicon and AlGaAs are 2.78 and 2.56, respectively. Figure 4A In the illustrated example implementation, to optimize mode switching between the first waveguide 402 and the second waveguide 428, a first inverted taper pair 436 is provided to optimize transient coupling of an optical signal propagating in the first waveguide 402 to the second waveguide 428, and a second inverted taper pair 418 is provided to optimize transient coupling of an optical signal propagating in the second waveguide 428 to the first waveguide 402.

[0070] For example, the first waveguide 402 may have a width W1 at the input junction 412 of the second waveguide 428 (e.g., closest to the optical signal coupled into the second waveguide 428) and a width W2 at the output junction 416 of the second waveguide 428 (e.g., closest to the optical signal coupled out of the second waveguide 428). In some embodiments, widths W1 and W2 may be the same or substantially similar in width. The second waveguide 428 may have a width W3 between the output junction 406 and the input junction 410 of the first waveguide 402. The output junction 406 may have a width W4, the input junction 410 may have a width W5, the output junction 416 may have a width W6, and the input junction 412 may have a width W7. In some embodiments, widths W1 and W2 may be the same or substantially similar in width, widths W4 and W5 may be the same or substantially similar in width, and widths W6 and W7 may be the same or substantially similar in width. In the above example implementation, W1 and W2 can be 500 nm; and W3 can be 600 nm (for ribbed waveguide 464, the total width is 4.6 μm); and W4 to W7 can be 200 nm.

[0071] Each pair of inverted cones 436 and 418 can be implemented as a corresponding mode converter (e.g., mode converter 436 and mode converter 418, respectively). The first pair of inverted cones 436 includes a cone that narrows the width W1 to the width W4 at the output junction 406 of the first waveguide 402 (similar to...). Figure 4A The second inverted cone pair 418 includes a cone (similar to the first waveguide 402) that widens the width W6 at the input junction 412 to the width W2. Figure 4A The cone 308) and the cone 444 that narrows the width W3 to the width W7 at the output junction 416 of the second waveguide 428. Figure 4A As shown, portions of the first waveguide 402 and the second waveguide 428 overlap in the Z-axis direction within a first tapered region 456 having a length L1 corresponding to the first inverted tapered pair 436, and portions of the first waveguide 402 and the second waveguide 428 overlap in the Z-axis direction within a second tapered region 458 having a length L2 corresponding to the second inverted tapered pair 418. In some embodiments, lengths L1 and L2 may be the same or substantially similar in width. According to some embodiments, such as Figure 4A As shown, plate 454 and bonding layer 452 may taper in a manner similar to the cone 442 of the first inverted cone pair 436, for example, within the first conical region 456. Similarly, plate 454 and bonding layer 452 may taper in a manner similar to the cone 444 of the second inverted cone pair 436, for example, within the second conical region 458.

[0072] The first inverted cone pair 436 facilitates the coupling of the fundamental TE mode between the first waveguide 402 and the second waveguide 428, while the second inverted cone pair 418 facilitates the coupling of the fundamental TE mode between the second waveguide 428 and the first waveguide 402. Optical coupling at each corresponding inverted cone pair 436 and 418 can be a result of the cones included therein. That is, the cone of the first inverted cone pair 436 can push the fundamental TE mode upwards into the second waveguide 428 so that it can be effectively coupled into the MRR 414, while the cone of the second inverted cone pair 418 can push the fundamental TE mode downstream into the first waveguide 402 and into the MZI 110. The first inverted cone pair 436 and the second inverted cone pair 418 can be designed such that they each satisfy certain performance characteristics. For example, the first inverted cone pair 436 and / or the second inverted cone pair 418 can be designed to be thermally insulating. As another example, the first inverted cone pair 436 and / or the second inverted cone pair 418 can be designed such that they are respectively coupled to a basic TE mode with very low loss and low back reflection. Mode converters implemented as having inverted cone pairs that are too short in length may suffer from high passive losses.

[0073] Figure 4D An example of mode conversion during the coupling of an optical signal from a first waveguide 402 to a second waveguide 428 within a first conical region 456 viewed along branch 122 is depicted. For example, Figure 4D The light intensities of the fundamental TE modes 430 and 432 along the X-axis direction are depicted. For the first inverted cone pair 436 with a cone length L1 of 60 μm, the coupling efficiency according to the embodiments disclosed herein can exceed 98%. Because the second waveguide 428 is wider, the modes in the second waveguide 428 appear less concentrated compared to the modes in the first waveguide 402.

[0074] according to Figures 4A-4D The second material in the example embodiment includes composition Al x Ga 1-x As, where x is at least greater than 0.1. In various embodiments, the value of x is greater than 0.2. In the example embodiment, the value of x is chosen to be 0.4. The band gap value of the second material can be based on the amount of aluminum in the composition. For example, Table 2 below provides Al x Ga 1-x Example values ​​of x and corresponding band gap values ​​for As.

[0075] Table 2

[0076] X Band gap (eV) 0 1.42 0.1 1.55 0.2 1.67 0.3 1.8 0.4 1.92 0.5 2

[0077] With an aluminum content of 0.4%, the second material has a bandgap of 1.92 EV, resulting in a TPA cutoff wavelength of 1290 nm. Therefore, the TPA effect at 1310 nm is negligible and does not provide nonlinearity. Optical loss is also low; for example, for an MRR with a radius of 12 μm formed from AlGaAs as described herein, the linear loss coefficient is as low as 60 dB / m. The quality factor of MRR 414 can be higher than 10. 6 Therefore, the nonlinear refractive index of AlGaAs is approximately 2.6 x 10⁻⁶. -17 m 2 / W, which is an order of magnitude larger than silicon, allows for a reduction in the input optical power to the activation device 100 to achieve nonlinear effects. A similar TPA effect can be achieved using an aluminum composition greater than 0.2.

[0078] Figure 5A-9C Graphical representations of various normalized nonlinear activation functions as a function of input optical power, according to embodiments disclosed herein, are depicted. The vertical axis of each graphical representation represents the output optical power, which is normalized to provide the normalized nonlinear activation function for each corresponding graph. Figure 5A , Figure 6A , Figure 7A , Figure 8A and Figure 9A A graphical representation of the normalized nonlinear activation function is depicted in the case where the MRR 114 of the activation device 100 includes a silicon waveguide. Figure 5B , Figure 6B , Figure 7B , Figure 8B and Figure 9B The case where the activation device 100 includes a second waveguide 328 and an MRR314 formed of Ta2O5 is depicted (as per [reference]). Figures 3A-3C The graphical representation of the normalized nonlinear activation function (described). Figure 5C , 6C 7C, 8C, and 9C depict the case where the activation device 100 includes a second waveguide 428 and an MRR 414 formed of AlGaAs (as described in the section on...). Figures 4A-4D The graphical representation of the normalized nonlinear activation function (described). Figures 5A-5C A graphical representation of the example radial basis activation function is depicted. Figures 6A-6C A graphical representation of the example ReLU activation function is depicted. Figures 7A-7C A graphical representation of the example inverse ReLU activation function is depicted. Figures 8A-8C A graphical representation of the example leaky ReLU activation function is depicted. Figures 9A-9C A graphical representation of an example quadratic activation function is depicted. Figure 5AIn each of -9D, r represents the power shunt ratio of MZC 106 (e.g., the ratio of power entering branch 124 of MZI 110 to power entering branch 122), which can be tuned using phase shift mechanism 108, and Δλ represents the detuning of the wavelength of the input optical signal from the resonant frequency of the corresponding MRR, which can be tuned using phase shift mechanism 112, and as mentioned above, the resonant frequency can be tuned using phase shift mechanism 116.

[0079] In the case of MRR 114 (e.g., silicon waveguides), the nonlinearity caused by the FCD effect is typically greater than that caused by the Kerr effect, meaning that the resonant frequency of silicon-formed MRRs will blue-shift at high power densities. However, in Figures 3A-4D In the case of MRR formed by Ta₂O₅ or AlGaAs, there is no FCD effect due to the presence of almost minimal free carriers, therefore the Kerr effect dominates the nonlinearity. Thus, as... Figure 5A-9C As shown, the relative wavelength detuning between the input optical signal and the resonant frequency of the corresponding MRR is set to be opposite to that between the Ta2O5 or AlGaAs waveguides.

[0080] In the case of an MRR 114 including a silicon waveguide, when the resonance of the MRR 114 is detuned at the origin of the graphical representation (e.g., relative detuning is towards the red end of the electromagnetic spectrum), a radial basis function-like behavior occurs (e.g., Figure 5A ). refer to Figure 5A The normalized output optical power (e.g., the normalized nonlinear activation function) initially increases linearly with the input optical power. As the resonant frequency of the MRR 114 blue-shifts towards the frequency of the input optical signal, the amplitude and phase of the MRR 114 undergo a sharp transition, and a decrease in output optical power is observed around the transition point. As the resonant frequency of the MRR 114 continues to blue-shift, less coupling occurs to the MRR 114, resulting in a sharp increase in output optical power with relatively small phase changes, such as... Figure 5A As shown in the image.

[0081] By further increasing the optical power of the input optical signal, the ReLU activation function can be obtained, such as... Figure 6A As shown in the diagram, the output optical power increases smoothly after the threshold input optical power, and the slope of the smooth increase region can be tuned by adjusting the splitting ratio r via the first phase shift mechanism 108 of the MZC 106.

[0082] When the resonant frequency is tuned to the frequency of the input optical signal, most of the optical signal is coupled to the MRR 114 (e.g., r is small) and the inverse ReLU is obtained, such as... Figure 7AAs shown in the diagram, the slope of the linear portion can be changed by altering the phase difference between the two branches of the MZI 110 via the second phase shift mechanism 112.

[0083] Leaky ReLU can be achieved by applying an additional π phase in a branch, such as... Figure 8A As shown in the diagram. For example, additional phase can be applied by tuning phase shift mechanism 112.

[0084] When the resonant frequency of MRR 114 is redshifted and r is small, a second-order nonlinear activation function can occur, corresponding to a sharp change in output optical power, such as... Figure 9A As shown in the image.

[0085] For MRR formed by Ta₂O₅ or AlGaAs waveguides, the above phenomenon operates in a manner similar to that described above but in the opposite direction. That is, for example, the above-described... Figure 5A , 6A The various tunes associated with 7A, 8A, and 9A for implementing various functions are opposite to the tunes of MRR 314 and / or MRR 414, and the ratios are reversed.

[0086] Due to the high nonlinearity and low loss of the second material according to the embodiments disclosed herein, the activation device 100 can switchably implement various nonlinear activation functions as described above, with significantly less input optical power compared to silicon structures. For example, due to the high nonlinearity and low loss of TA2O5, embodiments including Ta2O5 as the second material require one-third of the input optical power required by silicon structures, such as... Figure 5A As shown Figure 5B and 5C In comparison, similarly, due to the high nonlinearity and low loss of AlGaAs, implementations including AlGaAs as a second material require 1% of the input optical power compared to silicon structures. That is, Figure 5A The transition point occurs at an input optical power of 1.2e-3W, while Figure 5B The transition point occurs at 3e-4W and Figure 5C The transition point occurs at 1.2e-5W. A similar reduction in input optical power occurs at... Figure 6A-9C As shown, the choice between Ta2O5 and AlGaAs can depend on the desired optical power to achieve the desired nonlinear function (e.g., Figure 5A-9C As shown above), and specific projects and platforms. For example, as mentioned above, implementations including Ta2O5 require one-third of the input optical power compared to silicon structures, while implementations including AlGaAs require 1% of the optical power compared to silicon; however, Ta2O5 and AlGaAs have different integration methods on silicon platforms (as discussed above regarding...). Figures 3A-4D(As described). Therefore, the material has different uses because it can be manufactured using either Ta2O5 or AlGaAs for a given application and platform.

[0087] Therefore, the exemplary embodiments disclosed herein offer advantages for photonic neural applications because the power requirements in such networks are below 100 μW, which is lower than the power requirements of conventional all-optical activated devices. The lower input optical power level translates to lower power required from the input light source (e.g., a laser and / or a semiconductor optical amplifier (SOA)), thereby reducing power consumption and improving system linearity.

[0088] Therefore, the all-optical reconfigurable activation functions according to the embodiments disclosed herein offer several advantages. For example, a non-limiting advantage is that the integration of the SOI platform with Ta2O5 or AlGaAs results in lower input optical power requirements for generation and switching between various nonlinear activation functions compared to conventional methods. Furthermore, the embodiments disclosed herein allow for low-loss waveguides that can be designed for linear weight libraries and facilitate low-loss and high-efficiency neuromorphic photonic networks.

[0089] Another exemplary non-limiting advantage is that heterogeneous material-bonded AlGaAs platforms (e.g., Figures 4A-4D This could include a MOS capacitor structure formed of AlGaAs with a second waveguide 428 and a first waveguide 402, allowing control of the voltage bias applied between the materials to achieve phase shifting without any static power dissipation. This MOS structure is suitable not only for configurations (e.g., switching) between nonlinear functions but also for use in MZI 110. Light sources (e.g., lasers) can also be tightly integrated onto a single platform to minimize transition losses. While AlGaAs platforms may already include such MOS structures, some implementations using other second materials (such as Ta2O5) are suitable for utilizing MOS capacitor structures.

[0090] Another example of the non-limiting advantages is that the embodiments disclosed herein can be applied to existing silicon photonics platforms and are compatible with CMOS structures. Some passive and linear structures can be directly applied to SOI platforms.

[0091] Furthermore, the embodiments disclosed herein can generate activation functions based on the ultrafast Kerr effect, which enables high-speed operation. For example, high-speed nonlinear activation and / or variations between different activation functions (e.g., as described above regarding...). Figures 5A to 9C(As described above). The FCD effect (which limits processing speed due to the lifetime of free carriers) is negligible. Furthermore, since TPA-induced free carriers are negligible, thermally induced resonant shifts are also negligible. Therefore, the embodiments disclosed herein can operate at power intensity levels that are otherwise unattainable using silicon-based structures.

[0092] Phase tuning according to the embodiments disclosed herein can be achieved in many different ways. For example, throughout the phase shift mechanisms described herein, such as Figure 2 The phase shift mechanisms 108, 112, and / or 116 can be provided as any mechanism capable of inducing a phase shift in light propagating through the respective waveguide. For example, as indirectly mentioned above, phase shift mechanism 116 can be configured to tune the resonant frequency by inducing a change in the refractive index of the resonant structure (e.g., MMR 114) over a length, such as through carrier injection (e.g., charge accumulation), charge depletion, or changing the temperature of part or all of the resonator structure. As another example, phase shift mechanisms 108 and / or 112 can induce a phase shift within the respective branch by inducing a change in the refractive index of the waveguide over a length, such as through carrier injection (e.g., charge accumulation), charge depletion, or changing the temperature of part or all of the resonator structure.

[0093] In the first example, the phase tuning mechanism disclosed herein may include one or more heating / cooling elements (e.g., resistance heaters, etc.) operable to change the temperature of a waveguide (e.g., waveguide 302, waveguide 402, MRR 114, MRR 314, MRR 414, etc.). The heating element may be, for example, a resistor (e.g., a metal component) physically coupled to a portion or all of the corresponding waveguide. A current can then be applied to the resistor, which generates heat transferred to the corresponding waveguide, thereby causing a temperature change. Control of the current tunes the temperature, thereby tuning the phase and / or resonant frequency. A change in the phase or phase shift (Δφ) of the waveguide can be caused based on the following temperature variation:

[0094]

[0095] Where L H dn / dT is the length of the heating element coupled to the corresponding waveguide, dn / dT is the thermo-optic coefficient (e.g., indicating the change of refractive index with temperature) depending on the material forming the corresponding waveguide, ΔT is the temperature change of the corresponding waveguide, and λ0 is the free-space wavelength of light. Therefore, applying heat to or subtracting heat from the corresponding waveguide will cause a phase change.

[0096] Figure 10A and Figure 10BAnother example implementation of the phase-shifting mechanism according to the present disclosure is shown. Figure 10A and 10B An example hybrid MOS ring resonator optical modulator 1000 is shown, which can be implemented as a phase tuning mechanism, for example... Figure 2 One or more of phase shift mechanisms 108, 112 and 114. Figure 10A This is a perspective view of the optical modulator 1000, and Figure 10B It is along Figure 10A The cross-sectional view of the hybrid MOS optical modulator 1000 taken by line A-A' is shown in the figure.

[0097] The optical modulator 1000 includes an optical waveguide 1002, a cathode 1004 comprising a first material and formed therein, and an anode 1006 comprising a second material different from the first material and formed therein, the anode being adjacent to the cathode. A capacitor is defined between the anode and the cathode.

[0098] In some examples, substrate 1001 comprises an oxide grown on an underlying layer 1008. A silicon layer 1010 is formed on substrate 1001. Trench 1012 divides optical modulator 1000 into two portions 1014 and 1016, the first portion 1014 comprising an anode 1006. Optical waveguide 1002 is formed in anode 1006. Cathode 1004 is integrated into the second portion 1016. Optical waveguide 1002 may, for example, be briefly coupled to branch 118 in the case of phase shift mechanism 108, briefly coupled to branch 122 in the case of phase shift mechanism 112, and / or briefly coupled to a waveguide of MRR 114 in the case of phase shift mechanism 116. In various embodiments, cathode 1004 comprises a III-V group material layer as a first material. MOS capacitor 1024 is defined between cathode 1004 and anode 1006.

[0099] A dielectric 1018 is formed between the cathode 1004 and the anode 1006. The dielectric 1018 may be an electrically insulating material formed between the cathode 1004 and the anode 1006 of the MOS capacitor 1024, and polarizing the dielectric 1018 by applying an electric field can increase the surface charge of the MOS capacitor 1024 for a given electric field strength. The dielectric 1018 may be a natural oxide of the cathode or anode, or both, or it may be an external dielectric material, such as a high-k dielectric or polymer that can be formed by deposition, oxidation, wafer bonding, or other dielectric coating methods.

[0100] Cathode 1004 may comprise a negatively doped III-V group material (such as GaAs, AlGaAs, etc.), and anode 1006 may comprise a positively doped silicon. A cathode electrode 1020 is disposed on cathode 1004, and an anode electrode 1022 is disposed on anode 1006. When a voltage is applied between the electrodes, carrier accumulation, depletion, or reversal can occur around dielectric 1018. Since the capacitor region overlaps with the optical waveguide, changes in carrier concentration can lead to changes in the refractive index and propagation loss within waveguide 100. By applying a bias voltage between the electrodes, the refractive index can be modulated accordingly, thereby causing intensity modulation, phase shift modulation, and attenuation of the light.

[0101] When modulator 1000 is implemented as phase-shifting mechanism 108, light can be briefly coupled from the waveguide of branch 118 to optical waveguide 1002 (e.g., at least a portion of the light propagating in branch 118 is transferred to optical waveguide 1002). When modulator 1000 is implemented as phase-shifting mechanism 112, light can be briefly coupled from the waveguide of branch 122 to optical waveguide 1002 (e.g., at least a portion of the light propagating in the first waveguide of branch 118 is transferred to optical waveguide 1002). When modulator 1000 is implemented as phase-shifting mechanism 116, light can be briefly coupled from the second waveguide of branch 118 to optical waveguide 1002 (e.g., at least a portion of the light propagating in branch 118 is transferred to optical waveguide 1002). In each such embodiment, the light propagating through optical waveguide 1002 is modulated, attenuated, and phase-shifted based on the change in the refractive index of the waveguide mode caused by the bias of MOS capacitor 1024. A portion of the modulated and attenuated light can be briefly coupled from optical waveguide 1002 to another waveguide 1005. The other waveguide 1005 can be implemented as branch 118 (in the case where optical modulator 1000 is implemented as phase shift mechanism 108), as branch 122 (in the case where optical modulator 1000 is implemented as phase shift mechanism 112), and / or as a second waveguide 128 (in the case where optical modulator is implemented as phase shift mechanism 116).

[0102] For example, Figure 10BThis includes a DC power supply 1026. The DC power supply 1026 acts as a signal source and has a negative terminal connected to the cathode electrode 1020 and a positive terminal connected to the anode electrode 1022. This causes the migration of negative charges from the cathode 1004 toward the side of the optical waveguide 1002 adjacent to the cathode 1004, and the migration of positive charges (“holes”) from the anode 1006 to the opposite side of the waveguide 1002 (also referred to herein as accumulation mode). In other examples, the polarity of the DC power supply 1026 can be reversed. Reversing the polarity of the DC power supply 1026 causes the migration of negative charges from the waveguide 1002 toward the cathode electrode 1020, and the migration of holes from the waveguide 1002 toward the anode electrode 1022 (also referred to herein as depletion mode).

[0103] A MOS capacitor 1024 is formed at the boundary between the III-V material of the cathode 1004 and the underlying capacitor portion of intrinsic silicon or other group IV material. A thin layer of silicon and III-V oxide (e.g., dielectric 1018) naturally forms at this boundary and serves as the dielectric of the capacitor. In some examples, this thin layer has a nanometer-scale thickness, for example, a few nanometers thick. In some examples, no steps are required to stimulate the formation of dielectric 1018. In other examples, the formation of dielectric 1018 can be stimulated, for example, by increasing the temperature, exposing the material to an oxygen-rich environment, or by other suitable techniques.

[0104] As previously discussed, the MOS capacitor 1024 is formed inside the optical waveguide 1002 such that the charge carriers accumulated / depleted on either side of the capacitor dielectric have the effect of altering the refractive index and waveguide loss of the optical waveguide (e.g., loss or attenuation of signal power propagating in the waveform).

[0105] The MOS capacitor 1024 can operate in accumulation, depletion, or inversion modes (e.g., accumulating electrons at the dielectric layer in addition to the presence of holes). As discussed above, a DC voltage can be applied between the anode and cathode to cause the thin charge layer to accumulate, deplete, or invert on both sides of the dielectric layer 1018. The resulting change in free carrier density causes a change in the refractive index n of the optical waveguide 1002, which manifests as a change in the effective refractive index of the optical mode (Δn). eff The change or modulation amount of the effective refractive index (Δn) eff The change in optical loss (Δα) and the associated change in optical loss can be described as follows:

[0106]

[0107]

[0108] Where q is the charge applied to the cathode 1004 and anode 1006, c is the speed of light in vacuum, ε0 is the dielectric constant of free space, n is the refractive index of the material, and ΔN represents the change in carrier density, such that ΔN e ΔN represents the change in carrier density in terms of electrons. h This represents the change in carrier density with respect to holes, and m* represents the relative effective mass of electrons (m*). ce ) and the relative effective mass of holes (m* ch ), μ h μ represents hole mobility. e λ represents electron mobility, and λ0 is the free space wavelength.

[0109] The phase shift (Δφ) at the capacitor terminal depends on the voltage-induced Δn. eff The size of the device, its length L, and the wavelength λ. In the example, the optical phase shift can be calculated as Δφ = 2πΔn. eff Lλ. Therefore, the optical phase of the light within the optical waveguide 1002 can be based on Δn caused by the voltage. eff And shifting. In various examples, waveguide losses in silicon and group III-V materials can also change simultaneously with variations in carrier density, and the control of these variations can be used as optical attenuators. For example, the variation in waveguide loss can be controlled based on changes in carrier density, which can impart attenuation to the waveguide loss. Attenuated waveguide loss can be used to modulate signals.

[0110] like Figure 10A As shown, the optical modulator 1000 can be a ring resonator optical modulator. In this case, Figure 10B The trench 1012 shown is provided as an annular trench, which divides the optical modulator into a first portion 1014 and a second portion 1016, respectively. Similarly, the anode 1006 is provided as an annular anode in the second portion, while the cathode 1004, dielectric 1018, and silicon layer 1010 are cylindrical in the first portion. A MOS capacitor 1024 is defined across the boundary between the cathode and anode.

[0111] As described above, the depletion or accumulation of charge at the interface layer leads to a change in the free carrier density, altering the local refractive index of waveguide 1002. As described above, refer to... Figures 2 to 9CA change in the refractive index of waveguide 1002 can be used to induce a phase shift (e.g., phase offset) in the light propagating in waveguide 1002. When used as phase shift mechanism 108, a phase shift based on a voltage bias of MOSCAP 1024 can be used to tune MZC 106, for example, to tune the ratio of optical power supplied from the directional coupler to the respective branches of MZI 110 based on a controlled phase difference. When used as phase shift mechanism 112, a phase shift based on a voltage bias of MOSCAP 1024 can be used to tune the relative phase difference between the optical signal propagating in branch 124 and the optical signal propagating in branch 122. As mentioned above regarding Figures 5A to 9C As described, the relative phase difference between tuning branches 122 and 124 can configure activation device 100 between different activation functions. When used as phase shift mechanism 116, a phase shift based on the voltage bias of MOSCAP 1024 can be used to tune the resonant frequency of MRR 114. Controlled tuning of the resonant frequency of MRR 114 can be used to configure activation device 100 between different activation functions, as described above. Figures 5A to 9C As described.

[0112] Figure 11 Example computational components, according to various embodiments, are shown that can be used to operate nonlinear activation functions. Reference is now made to... Figure 11 The computing component 1100 can be, for example, a server computer, a controller, or any other similar computing component capable of processing data. Figure 11 In an example implementation, computing component 1100 includes hardware processor 1102 and machine-readable storage medium 1104.

[0113] The hardware processor 1102 may be one or more central processing units (CPUs), semiconductor-based microprocessors, and / or other hardware devices suitable for retrieving and executing instructions stored in the machine-readable storage medium 1104. The hardware processor 1102 may acquire, decode, and execute instructions such as instructions 1106-1110 to control processes or operations for manipulating nonlinear activation functions, such as combining... Figure 2-4D As an alternative to or supplement to retrieving and executing instructions, hardware processor 1102 may include one or more electronic circuits, such as field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), or other electronic circuits, which include electronic components for performing one or more instructions.

[0114] Machine-readable storage media such as machine-readable storage medium 1104 can be any electronic, magnetic, optical, or other physical storage device that includes or stores executable instructions. Therefore, machine-readable storage medium 1104 can be, for example, random access memory (RAM), non-volatile RAM (NVRAM), electrically erasable programmable read-only memory (EEPROM), storage devices, optical discs, etc. In some embodiments, machine-readable storage medium 1104 can be a non-transitory storage medium, wherein the term "non-transitory" does not cover transient propagation signals. As described in detail below, machine-readable storage medium 1104 can be encoded with executable instructions (e.g., instructions 1106-1110).

[0115] Hardware processor 1102 can execute instructions 1106 to control a first phase shift mechanism to adjust the first bias of the MZI to tune the phase difference between the branches of the MZI. For example, the MZI can be implemented as MZI 110 and the first phase shift mechanism can be implemented as... Figure 2 The phase shift mechanism 112. As described above, the phase shift mechanism 112 can be controlled to tune the phase difference between branch 122 and branch 124, thereby converting the nonlinear phase into a nonlinear response transmitted via the output of the MZI.

[0116] Hardware processor 1102 can execute instructions 1108 to control a second phase shift mechanism of the MRR coupled to one of the branches of the MZI, to adjust a second bias of the MRR so that the nonlinear activation device operates at approximately the resonant frequency of the MRR. For example, the MRR can be implemented as follows: Figure 2-4D The MRR is one of 114, 314, and / or 414, and the second phase shift mechanism is implemented as phase shift mechanism 116. As described above, phase shift mechanism 116 can be controlled to tune the resonant frequency of the MRR, thereby providing the nonlinearity converted by phase shift mechanism 112. As described above, the MRR includes the features described above regarding... Figure 2-4D The waveguide is formed of a second material. For example, the second material may have a band gap greater than 1.9 eV and a density greater than 3 x 10⁻⁶ eV. -20 Materials with nonlinear refractive index and linear loss coefficient less than 200 dB / m.

[0117] Hardware processor 1102 can execute instructions 1110 to control the third phase shift mechanism of the MZC coupled to the MZI, adjusting the third bias of the MZC to tune the amplitude of the branches of the MZI relative to each other. For example, the MZC can be implemented as MZC 106 having a phase shift mechanism 108 coupled to one of its branches. MZC 106 serves as a tunable directional coupler tuned based on phase shift mechanism 108. As described above, phase shift mechanism 108 provides tuning of the splitting ratio of the optical power supplied to each branch of the MZI 100. Therefore, the amplitude of the optical power supplied to each branch of the MZI 110 can be balanced by tuning phase shift mechanism 108.

[0118] Therefore, as referenced above Figure 5A-9C As described, the first, second, and third biases can be controlled individually and independently by adjusting the phase and optical power within the all-optical activation device, thereby achieving a desired activation function. For example, phase shift mechanisms 108, 112, and 116 can be controlled to set the first, second, and third biases as a first condition to achieve a first activation function (e.g., Figure 5A (Radial base class functions). Then one or more of phase shift mechanisms 108, 112, and 116 can be tuned to adjust one or more of the first, second, and third biases to the second condition to achieve the second activation function (e.g., by tuning the ReLU activation function of the optical power of the input optical signal, such as...). Figure 6A As shown; the resonant frequency is tuned to achieve the inverse ReLU activation function, as described in conjunction with Figures 5a-9c.

[0119] Figure 12 A block diagram is depicted illustrating an example computer system 1200 in which various embodiments described herein may be implemented. The computer system 1200 may be implemented as, for example, one or more of control phase-shift mechanisms 108, 112, and / or 116, and includes a bus 1202 or other communication mechanism for communicating information, and one or more hardware processors 1204 coupled to the bus 1202 for processing information. The hardware processors 1204 may be, for example, one or more general-purpose microprocessors.

[0120] Computer system 1200 also includes main memory 1206, such as random access memory (RAM), cache, and / or other dynamic storage devices, coupled to bus 1202, for storing information and instructions (e.g., instructions 1106-1110) to be executed by processor 1204. Main memory 1206 can also be used to store temporary variables or other intermediate information during the execution of instructions by processor 1204. When stored in a storage medium accessible to processor 1204, such instructions make computer system 1200 a dedicated machine customized to perform the operations specified in the instructions.

[0121] Computer system 1200 also includes a read-only memory (ROM) 1208 or other static storage device coupled to bus 1202 for storing static information and instructions for processor 1204. Storage devices 1210 coupled to bus 1202 for storing information and instructions are provided, such as disks, optical discs, or USB thumb drives (flash drives).

[0122] Computer system 1200 may be coupled to display 1212, such as a liquid crystal display (LCD) (or touchscreen), via bus 1202 for displaying information to a computer user. Input device 1214, including alphanumeric keys and other keys, is coupled to bus 1202 for transmitting information and command selections to processor 120. Another type of user input device is cursor control 1216, such as a mouse, trackball, or arrow keys, for transmitting directional information and command selections to processor 1204 and for controlling cursor movement on display 1212. In some embodiments, the same directional information and command selections as cursor control may be received via touch on a touchscreen without a cursor.

[0123] The computing system 1200 may include a user interface module to implement a GUI, which may be stored as executable software code executed by the computing device in a mass storage device. This module and other modules may include, for example, components such as software components, object-oriented software components, class components and task components, processes, functions, properties, procedures, subroutines, program code segments, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables.

[0124] Generally, terms such as “component,” “engine,” “system,” “database,” and “data storage” as used herein can refer to logic embodied in hardware or firmware, or to a set of software instructions written in a programming language (such as, for example, Java, C, or C++) that may have entry and exit points. Software components can be compiled and linked into an executable program, installed in a dynamic link library, or written in an interpreted programming language such as, for example, BASIC, Perl, or Python. It is understood that software components can be invoked from other components or from themselves, and / or can be invoked in response to detected events or interrupts. Software components configured to execute on a computing device can be provided on computer-readable media, such as optical discs, digital video discs, flash drives, disks, or any other tangible media, or as digital downloads (and can be initially stored in a compressed or installable format that needs to be installed, decompressed, or decrypted prior to execution). Such software code can be stored, partially or entirely, on a memory device executing the computing device for execution by the computing device. Software instructions can be embedded in firmware, such as EPROM. It will be further understood that the hardware components may consist of connected logic units (such as gates and flip-flops), and / or may consist of programmable units (such as programmable gate arrays or processors).

[0125] Computer system 1200 may implement the techniques described herein using custom hardwired logic, one or more ASICs or FPGAs, firmware, and / or program logic, which, when combined with the computer system, enables computer system 1200 to be a special-purpose machine or to be programmed as such. According to one embodiment, the techniques described herein are executed by computer system 1200 in response to processor 1204 executing one or more sequences of one or more instructions contained in main memory 1206. Such instructions may be read into main memory 1206 from another storage medium, such as storage device 1210. Execution of the sequence of instructions contained in main memory 1206 causes processor 1204 to perform the process steps described herein. In alternative embodiments, hardwired circuitry may be used in place of or in combination with software instructions.

[0126] As used herein, the term "non-transitory media" and similar terms refer to any medium that stores data and / or instructions that enable a machine to operate in a particular manner. Such non-transitory media can include non-volatile media and / or volatile media. Non-volatile media include, for example, optical discs or magnetic disks, such as storage device 1210. Volatile media include dynamic memory, such as main memory 1206. Common forms of non-transitory media include, for example, floppy disks, flexible disks, hard disks, solid-state drives, magnetic tape or any other magnetic data storage media, CD-ROMs, any other optical data storage media, any physical media with a perforated pattern, RAM, PROMs and EPROMs, FLASH-EPROMs, NVRAMs, any other memory chips or cartridges, and the same networking versions.

[0127] Non-transitory media differ from transmission media but can be used in conjunction with them. Transmission media participate in the transmission of information between non-transitory media. For example, transmission media include coaxial cables, copper wires, and optical fibers, including the conductors constituting bus 1202. Transmission media can also take the form of sound waves or light waves, such as those generated during radio wave and infrared data communication.

[0128] Computer system 1200 also includes a communication interface 1218 coupled to bus 1202. Network interface 1218 provides bidirectional data communication coupled to one or more network links connected to one or more local networks. For example, communication interface 1218 may be an Integrated Services Digital Network (ISDN) card, a cable modem, a satellite modem, or a modem providing a data communication connection to a corresponding type of telephone line. As another example, network interface 1218 may be a local area network (LAN) card providing a data communication connection to a compatible LAN (or a WAN component communicating with a WAN). Wireless links may also be implemented. In any such implementation, network interface 1218 transmits and receives electrical, electromagnetic, or optical signals carrying streams of digital data representing various types of information.

[0129] Network links typically provide data communication to other data devices via one or more networks. For example, a network link may provide a connection to a host computer or to a data device operated by an Internet Service Provider (ISP) via a local network. The ISP, in turn, provides data communication services through a global packet data communication network now commonly referred to as the "Internet." Both local networks and the Internet use electrical, electromagnetic, or optical signals that carry digital data streams. Signals through various networks and on network links, as well as signals through communication interface 1218 (which carry digital data to and from computer system 1200), are example forms of transmission media.

[0130] Computer system 1200 can send messages and receive data, including program code, via a network, network link, and communication interface 1218. In the Internet example, the server can send requested code for an application via the Internet, ISP, local network, and communication interface 1218.

[0131] The received code may be executed by processor 1204 when it is received, and / or stored in storage device 1210 or other non-volatile memory for later execution.

[0132] Each of the processes, methods, and algorithms described in the preceding sections can be embodied in code components executed by one or more computer systems or computer processors, including computer hardware, and can be fully or partially automated. One or more computer systems or computer processors can also operate to support the execution of related operations in a “cloud computing” environment or as “Software as a Service” (SaaS). Processes and algorithms can be implemented, partially or entirely, in dedicated circuitry. The various features and processes described above can be used independently of each other or can be combined in various ways. Different combinations and sub-combinations are intended to fall within the scope of this disclosure, and certain method or process blocks may be omitted in some embodiments. The methods and processes described herein are not limited to any particular sequence, and associated blocks or states can be executed in other suitable sequences, or can be executed in parallel or in some other way. Blocks or states can be added to or removed from the disclosed example embodiments. The execution of certain operations or processes can be distributed among computer systems or computer processors, residing not only within a single machine but also deployed across multiple machines.

[0133] As used herein, the circuit can be implemented using any form of hardware, software, or a combination thereof. For example, one or more processors, controllers, ASICs, PLAs, PALs, CPLDs, FPGAs, logic components, software routines, or other mechanisms can be used to construct the circuit. In implementation, the various circuits described herein can be implemented as discrete circuits, or the described functions and features can be shared partially or wholly among one or more circuits. Although various features or functional elements can be described or claimed as separate circuits, these features and functions can be shared among one or more common circuits, and such description does not require or imply the need for separate circuits to implement such features or functions. In the case where the circuit is implemented wholly or partially using software, such software can be implemented to operate in conjunction with a computing or processing system (such as computer system 1200) capable of performing the functions described herein.

[0134] As used herein, the term “or” can be interpreted in an inclusive or exclusive sense. Furthermore, descriptions of resources, operations, or structures in the singular should not be construed as excluding the plural form. Unless expressly stated otherwise, or otherwise understood in the context in which they are used, conditional language such as “may,” “can,” “may,” or “may,” among others, is generally intended to convey that certain embodiments include certain features, elements, and / or steps that are not included in other embodiments.

[0135] Unless otherwise expressly stated, the terms and phrases used herein, and their variations thereof, should be interpreted as open-ended rather than restrictive. Adjectives and terms with similar meanings such as “regular,” “traditional,” “usual,” “standard,” “known,” and similar expressions should not be interpreted as limiting the described item to a given time period or to items available at a given time, but rather should be interpreted as encompassing regular, traditional, usual, or standard techniques that may be available or known at any time now or in the future. In some instances, the presence of extended words and phrases such as “one or more,” “at least,” “but not limited to,” or other similar phrases should not be interpreted as anticipating or requiring a narrower application in situations where these extended phrases might not be present.

Claims

1. An optical device, comprising: A directional coupler, the directional coupler including a first phase shift mechanism; An interferometer, coupled to the orientation coupler, the interferometer comprising: The first branch includes a first waveguide formed of a first material and a second waveguide formed of a second material different from the first material. The second branch, comprising a third waveguide formed of the first material, and The second phase shift mechanism coupled to the first waveguide; and A microring resonator, coupled to the second waveguide, the microring resonator being formed of the second material and including a third phase-shifting mechanism. The first phase-shift mechanism, the second phase-shift mechanism, and the third phase-shift mechanism are configured to control the bias of the optical device to achieve a desired activation function. The second material includes at least one of the following: a band gap exceeding 1.9 eV; a band gap greater than 3 x 10⁻⁶ eV. -20 m 2 The nonlinear refractive index is / W, and the linear loss coefficient is less than 200dB / m.

2. The optical device according to claim 1, wherein, The second material is aluminum gallium arsenide (AlGaAs).

3. The optical device according to claim 1, wherein, The second material is tantalum pentoxide (Ta2O5).

4. The optical device according to claim 1, wherein, The ratio of the nonlinear refractive index to the linear loss coefficient of the second material is greater than 3 x 10⁻⁶. -20 .

5. The optical device according to claim 1, wherein, The first waveguide includes a junction end coupled to the junction end of the second waveguide, and the optical device further includes: A first inverted cone pair, comprising a first cone at the junction of the first waveguide and a second cone at the junction of the second waveguide, wherein the first cone is in the opposite direction to the second cone.

6. The optical device according to claim 1, wherein, The third phase shift mechanism includes at least one of the following: a heterostructure metal-oxide-semiconductor (MOS) phase shifter and a heater.

7. The optical device according to claim 1, wherein, At least the third phase shift mechanism includes a heterostructure metal-oxide-semiconductor (MOS) capacitor.

8. The optical device according to claim 1, wherein, The directional coupler is a Mach-Zern coupler and the interferometer is a Mach-Zern interferometer.

9. A nonlinear activation device, comprising: Mach-Zende coupler MZC, the MZC including a first phase shift mechanism; A Mach-Zenith interferometer (MZI) coupled to the MZC, the MZI comprising a first branch, the first branch comprising a first waveguide formed of a first material and a second waveguide formed of a second material different from the first material, and a second phase shift mechanism coupled to the first waveguide; as well as A microring resonator (MRR) is coupled to the second waveguide, and the microring resonator is formed of the second material. The first phase-shift mechanism and the second phase-shift mechanism are configured to control the bias of the nonlinear activation device to switchably achieve the desired activation function. The second material includes at least one of the following: a band gap exceeding 1.9 eV; a band gap greater than 3 x 10⁻⁶ eV. -20 m 2 The nonlinear refractive index is / W, and the linear loss coefficient is less than 200dB / m.

10. The nonlinear activation device according to claim 9, wherein, The ratio of the nonlinear refractive index to the linear loss coefficient of the second material is greater than 5 x 10. -20 .

11. The nonlinear activation device according to claim 9, wherein, The second material is aluminum gallium arsenide (AlGaAs).

12. The nonlinear activation device according to claim 9, wherein, The second material is tantalum pentoxide (Ta2O5).

13. The nonlinear activation device according to claim 9, wherein, The first waveguide includes a first junction end coupled to a second junction end of the second waveguide, and the nonlinear activation device further includes: a first inverted cone pair, the first inverted cone pair including a first cone at the first junction end of the first waveguide and a second cone at the second junction end of the second waveguide, wherein the first cone is in the opposite direction to the second cone.

14. The nonlinear activation device according to claim 13, wherein, The first waveguide includes a third junction coupled to a fourth junction of the second waveguide, and the nonlinear activation device further includes a second inverted cone pair, the second inverted cone pair including a third cone at the third junction of the first waveguide and a fourth cone at the fourth junction of the second waveguide, wherein the third cone is opposite to the fourth cone and opposite to the first cone.

15. A method for operating a nonlinear activation device, the method comprising: A first bias of the MZI, included in the nonlinear activation device, is adjusted by controlling a first phase shift mechanism of the Mach-Zernd interferometer (MZI) to tune the phase difference between the branches of the MZI, wherein the first phase shift mechanism is coupled to a first waveguide of the first branch of the MZI. The second bias of the MRR, coupled to the first branch of the MZI, is adjusted by controlling the second phase shift mechanism of the MRR, such that the nonlinear activation device operates at approximately the resonant frequency of the MRR, which is formed of a second material comprising at least one of the following: a bandgap exceeding 1.9 eV; a bandgap greater than 3 x 10⁻⁶ eV; and a bandgap greater than 3 x 10⁻⁶ eV. -20 m 2 The nonlinear refractive index is / W, and the linear loss coefficient is less than 200dB / m; as well as The third bias of the MZC coupled to the MZI is adjusted by controlling the third phase shift mechanism of the Mach-Zenit coupler MZC to tune the amplitude of the branches of the MZI relative to each other. Specifically, the first bias, the second bias, and the third bias are controlled to achieve the desired activation function, and The first waveguide is formed of a first material, and the second waveguide is formed of a second material different from the first material.

16. The method according to claim 15, wherein, Each of the first phase shift mechanism, the second phase shift mechanism, and the third phase shift mechanism includes one of a heterostructure metal-oxide-semiconductor (MOS) phase shifter or a heater.

17. The method according to claim 15, wherein, The second material of the MRR is one of aluminum gallium arsenide (AlGaAs) and tantalum pentoxide (Ta2O5).

Images