Skip to content
BY 4.0 license Open Access Published by De Gruyter May 25, 2022

Programmable chalcogenide-based all-optical deep neural networks

  • Ting Yu Teo ORCID logo EMAIL logo , Xiaoxuan Ma , Ernest Pastor , Hao Wang ORCID logo , Jonathan K. George , Joel K. W. Yang , Simon Wall ORCID logo , Mario Miscuglio , Robert E. Simpson ORCID logo EMAIL logo and Volker J. Sorger EMAIL logo
From the journal Nanophotonics


We demonstrate a passive all-chalcogenide all-optical perceptron scheme. The network’s nonlinear activation function (NLAF) relies on the nonlinear response of Ge2Sb2Te5 to femtosecond laser pulses. We measured the sub-picosecond time-resolved optical constants of Ge2Sb2Te5 at a wavelength of 1500 nm and used them to design a high-speed Ge2Sb2Te5-tuned microring resonator all-optical NLAF. The NLAF had a sigmoidal response when subjected to different laser fluence excitation and had a dynamic range of −9.7 dB. The perceptron’s waveguide material was AlN because it allowed efficient heat dissipation during laser switching. A two-temperature analysis revealed that the operating speed of the NLAF is 1 ns. The percepton’s nonvolatile weights were set using low-loss Sb2S3-tuned Mach Zehnder interferometers (MZIs). A three-layer deep neural network model was used to test the feasibility of the network scheme and a maximum training accuracy of 94.5% was obtained. We conclude that combining Sb2S3-programmed MZI weights with the nonlinear response of Ge2Sb2Te5 to femtosecond pulses is sufficient to perform energy-efficient all-optical neural classifications at rates greater than 1 GHz.

1 Introduction

With the advent of artificial intelligence, computers are tasked to mimic the biological brain and perform data-centric tasks, like image and speech recognition. However, implementing and training artificial neural networks (ANN) are computationally intensive when large amounts of data are being processed. Current computing technologies are not optimized to efficiently perform data-centric tasks. It is estimated that by 2040, computers will require more energy than we could generate to process the data created [1]. Hence, faster and more energy-efficient computing hardware must be developed to cope with this demand.

New computing paradigms have emerged in recent years to enhance computer performance. One promising solution transforms the virtual ANN machine learning algorithm into hardware. Neural-inspired application specific hardware devices can accelerate multiply-and-accumulate (MAC) operations, which constitutes the weighting functionality of an ANN. On-chip neural-inspired weights have been extensively developed [2], [3], [4]. These devices are usually implemented in the optical domain to avoid electrical circuit inefficiencies like lossy electrical interconnects and capacitive charging [5].

Despite the recent advances in neural-inspired optical weighting devices, the other necessary ANN components are still reliant on current computing technologies. This includes the nonlinear activation function (NLAF) [2], [3], [4], which is essential for network training and the decision mapping process. Data shuttling between the hardware weights and virtual ANN components will compromise network speed and energy usage. There have been efforts to develop the NLAF device. However, current hardware NLAF components require further optimization as they have not been demonstrated on a device level [6], require electrical to optical conversion [7], are inefficient in terms of speed or/and energy usage [8, 9], or have complicated fabrication processes [10].

To develop a full hardware neural network, the next design iteration involves efficiently integrating the NLAF with the weights. Integrating both components together is critical as they constitute the perceptron, which is the basic building block of a neural network. Devices within the perceptron must exhibit both nonvolatile and volatile tunability. The weighting components must be nonvolatile to retain programmed weights when performing MAC operations. Meanwhile, the NLAF device must be volatile as it needs to retain its original mapping function when processing each incoming signal.

Using chalcogenide phase change material (PCM)-based reconfigurable photonic devices to develop an all-optical perceptron is desirable as conventional photonic waveguide devices can now become tunable and perform switching operations. Although structural phase transitions in chalcogenides have already been proposed to tune the optical neural networks weighted interconnections [3, 11], they have not been studied for performing nonlinear activation. Chalcogenides are known to exhibit a range of nonlinear optical effects, especially third order nonlinearity, χ 3 [12]. These nonlinearities could be used to implement the NLAF in an all-optical perceptron.

Besides the third order nonlinearity, χ 3 , the highly nonlinear volatile change to the optical constants of crystalline Ge2Sb2Te5 , which was previously reported, could be important for creating NLAFs [13, 14]. The p-orbital electrons in some crystalline chalcogenides, such as Ge2Sb2Te5, can be delocalized and therefore highly polarizable, which enhances the material’s dielectric permittivity. This was demonstrated by Shportko et al. where there was a large increase in electronic polarizability when the PCM changed from the amorphous to crystalline state [15]. Several studies suggest that this permittivity enhancement can be temporarily disrupted by an external stimulus, such as short laser or electrical pulses [13, 16], [17], [18]. We hypothesized that this high-speed volatile change in dielectric permittivity could be exploited to generate a NLAF. Moreover, by combining structural transitions to set the network’s nonvolatile weights with the Ge2Sb2Te5 nonlinear activation function, we further hypothesized that feed-forward neural network inference operations would be possible without converting the optical signal into the electrical domain.

Herein, we designed, modeled, and optimized a low-loss all-optical chalcogenide perceptron that can recognize images using a convolutional neural network approach. In the proposed all-chalcogenide optical perceptron model, both the weights and the NLAF consist of chalcogenide reconfigurable devices. The perceptron device consists of a low-loss Sb2S3-tuned MZI, which is used to set the network weights, and a sigmoidal NLAF using a crystalline Ge2Sb2Te5-tuned ring resonator. We experimentally show volatile switching of crystalline Ge2Sb2Te5, with the optical properties being tuned within 1 ps when excited with a femtosecond laser pulse. For the first time, this transient optical response of Ge2Sb2Te5 was modeled on a photonic device to implement a sigmoidal NLAF. Based on our simulation model, the sigmoidal NLAF has a dynamic range of 9.7 dB. Using a two-temperature model to describe how femtosecond excitation can influence the NLAF transmission, we show that the perceptron can have an overall computation speed of more than 1 GHz in a feedforward neural network scenario. An optical neural network was modeled using the proposed all-chalcogenide-tuned perceptron design. Network performance accuracy of up to 94.5% was achieved when used to infer characters from the MNIST dataset.

2 Materials and methods

2.1 Ge2Sb2Te5 material fabrication

Amorphous Ge2Sb2Te5 thin films were deposited on glass substrates by radio frequency magnetron sputtering using an AJA Orion 5 sputtering system with a base pressure of 2.5 × 10−7 Torr. The Ge2Sb2Te5 material comes from a commercially purchased target with diameter of 50.8 mm and purity of 99.9%. The sputtering process took place in an argon environment at a pressure of 3.7 × 10−3 Torr. The RF power was set to 30 W, which resulted in a deposition rate of 0.093 Å/s. The deposition rate was calibrated using step profilometry of films deposited for a fixed time.

To crystallize the amorphous Ge2Sb2Te5 thin film, we first determined the phase transition temperature. This was done by recording the change in optical reflectivity as we heated the material. The first abrupt change in reflectivity of the thin film indicates an amorphous to face-centered cubic (FCC) structural phase transition [19]. The as-deposited amorphous Ge2Sb2Te5 thin film was heated to 250 °C, at a heating rate of 5 °C/min in a Linkam heating stage (Linkam T95-HT). To prevent oxidation, Ar gas at 4 SCCM was supplied to the heating enclosure. The material reflectivity was recorded each time the heating stage temperature increased by 1 °C. To ensure consistency, this sample was from the same sputtering batch as the sample to be used in the femtosecond laser switching experiment. The phase transition temperature was found to be at 176 °C. Upon determining the phase transition temperature, we then crystallize the actual amorphous Ge2Sb2Te5 sample used in the femtosecond laser switching experiment. To ensure that the material was fully crystallized, we heated the sample to 183 °C with a heating ramp rate of 5 °C/s and a hold time of 30 min. Similarly, an Ar gas flow of 4 SCCM was supplied to the heating enclosure to prevent oxidation. The reflectivity measurements of the Ge2Sb2Te5 film can be found in the Supplementary Material Section 1.

2.2 Time-resolved femtosecond switching experiment

The pump–probe transient transmission measurements were performed on the crystalline Ge2Sb2Te5 at a wavelength of 1500 nm after pumping with an 800 nm pulse of 35 fs duration. A commercial regenerative 5 kHz Ti:Sapphire amplifier (Coherent) generated the 800 nm pump pulses, which were split and sent to the sample and to a commercial OPA (Light conversion TOPAS) to generate the probe pulse (1500 nm). To ensure a uniform excitation area, the pump spot (FWHM 180 × 100 µm) was significantly larger than the probe. A mechanical chopper at 500 Hz was used to modulate the pump and the changes in reflectance, Δ R exp R exp , and in transmittance, Δ T exp T exp , with respect to time were simultaneously recorded using two InGaAs detectors. Note, subscript “exp” represents the experimental measurements. We derived the corresponding changes in the optical constants (refractive index n and dielectric function k) from the measured Δ R exp R exp and Δ T exp T exp using a procedure previously reported [13]. In brief, the procedure involved calculating the changes in transmission:

(1) Δ T calc T calc ( n , d , λ ) = T ( n , d , λ ) T ( n 0 , d , λ ) 1

and reflection:

(2) Δ R calc R calc ( n , d , λ ) = R ( n , d , λ ) R ( n 0 , d , λ ) 1

for the multilayer structure (Ge2Sb2Te5/SiO2 in our case) using a transfer matrix method described by Born et al. [20] at a given wavelength, λ, thickness, d. Moreover, n 0 = 6.5 + 1.2i and n′ are the initial and perturbed complex refractive indices, respectively. Subsequently, we fitted the change in transmittance and reflectance, Δ T calc T calc and Δ R calc R calc , to the measured experimental data, Δ T exp T exp and Δ R exp R exp , and obtained the values of n that gave the best fit to both datasets simultaneously. This allowed us to obtain a unique solution for the optical constant values of Ge2Sb2Te5. The analysis was performed at each probe delay. Multiple reflections in the SiO2 substrate were ignored. The effects of small variations in the Ge2Sb2Te5 thicknesses (24, 27 and 30 nm, respectively) were found to only give a small error in the retrieved refractive index. To account for the change in material thickness upon crystallization [21], our analysis and simulation models used the mean value of the optical constants derived from the three thickness values.

2.3 Waveguide design and modeling

2.3.1 Ring resonator

The Microring Resonator (MRR) NLAF spectrum was calculated using the power transfer function [22]:

(3) | S 21 | = υ 2 2 υ ξ cos ϕ + ξ 2 1 2 υ ξ cos ϕ + ( υ ξ ) 2

where υ is the self-coupling coefficient, ϕ is the phase shift and ξ is the amplitude attenuation coefficient for one round trip in the ring.

For the Ge2Sb2Te5-tuned ring resonator, ϕ and ξ can be expressed as [23]:

(4) ϕ = 2 π λ n eff , AlN ( L x ) + 2 π λ n eff , GST ( x )

(5) ξ = e α AlN ( L x ) + α GST ( x ) 2

where λ is the wavelength, n eff is the effective refractive index, α is the absorption coefficient, L is the ring length and x is the length of the Ge2Sb2Te5.

The n eff values and α GST were calculated using the eigenmode solver in Lumerical Mode Solution. The corresponding n eff values and mode profile of the AlN waveguide and Ge2Sb2Te5-based AlN waveguide can be found in the Supplementary Material Section 2. Optical constants used in the simulation model were as follows: AlN waveguide and SiO2 substrate were from references [24, 25] and Ge2Sb2Te5 was from Figure 3D and E. The α AlN value was set to be 0.8 dB/cm as reported in reference [26].

υ is expressed as:

(6) υ 2 + m 2 = 1 loss

where m is the cross-coupling coefficient and the loss represents the losses in the coupling section.

The AlN ring resonator spectrum reported in reference [26] was replicated using Eqs. (3)(5) to estimate the loss value in the coupling section expressed in Eq. (6). m was derived by modeling the coupling region on Lumerical and performing 3D Finite Difference Time Domain (3D FDTD) calculations.

2.3.2 Directional coupler

The design of the Sb2S3 directional coupler for the MZI weights was reported in references [27, 28]. The optical constants used for the waveguide and substrate were identical to the MRR model and the Sb2S3 optical constants were from reference [28]. The optimized dimensions of the directional coupler and the mode profile of the AlN waveguide and Sb2S3-based AlN waveguide can be found in the Supplementary Material Section 3.

3 Results and discussion

3.1 All-optical perceptron design

The proposed all-chalcogenide perceptron is shown in Figure 1A. The multiplication operation is performed by a photonic programmable phase-change array (P3A) as shown in Figure 1B, which consist of a network of reprogrammable MZI. In the MZI, one of the coupling waveguides in the input directional coupler arm contains a strip of PCM; as shown in the inset of Figure 1B. The amorphous to crystal ratio of the PCM strip is tuned to represent different weight values [28]. The resultant output signal from each MZI is then summed together when the waveguides converge to one point. The optical output of the MAC operation is then amplified to program the PCM on the NLAF component. Amplification is done in the optical domain using either an on-chip semiconductor optical amplifiers (SOAs) or Erbium Doped Fiber Amplifiers (EDFAs). Figure 1C shows the NLAF array made from MRRs. The final output of the perceptron is an optical signal that propagates through the NLAF component as the PCM NLAF is simultaneously programmed by the weights. Note that the optical input of the NLAF, as indicated by the green arrow in Figure 1C, is a separate probe signal from the MAC output.

Figure 1: 
All-chalcogenide optical perceptron model. (A) Schematic diagram of the perceptron. (B) P3A consisting of a network of reprogrammable MZI weights. Reconfigurability is introduced by the Sb2S3 - tuned directional coupler arm. (C) Ge2Sb2Te5-tuned MRR NLAF.
Figure 1:

All-chalcogenide optical perceptron model. (A) Schematic diagram of the perceptron. (B) P3A consisting of a network of reprogrammable MZI weights. Reconfigurability is introduced by the Sb2S3 - tuned directional coupler arm. (C) Ge2Sb2Te5-tuned MRR NLAF.

Optimizing the PCM for each component is critical for efficient network performance. The PCM in the weighting component must retain its optical property. It must also be low loss to allow network scalability. Common PCM data storage materials, such as Ge2Sb2Te5, are too absorbing in both the visible and infrared. Recently, a new generation of wide bandgap PCMs were introduced [29, 30] and their optical losses are lower than the Te-based PCM compounds. Sb2S3 was reported to have the largest bandgap in the visible and N-IR amongst the emerging PCMs [28, 31], which corresponds to the lowest optical absorption. We recently showed that Sb2S3 is the most promising candidate material for programming optical directional couplers [28] due to its low-loss property. Moreover, Sb2S3 can be programmed with different degrees of crystallinity by partial amorphization [32]. For these reasons, we chose Sb2S3 as the PCM to tune the weights of the P3A.

The PCM in the NLAF component must exhibit a nonlinear change in optical properties and volatile tunability. Whilst chalcogenides tend to exhibit a range of nonlinear optical effects, for instance χ 3 , it may be challenging to implement them on nanophotonic platforms. This is because long optical path lengths are required for phase accumulation. In contrast, Tellurides, in just tens of nm thickness, tend to show nonlinear changes in optical properties when the delocalized p-orbital bonds in the crystalline structure are disrupted. The delocalized p-orbital bonds enhance the optical matrix elements. This was attributed to resonant bonding [15] but more recently distinct difference to textbook resonant bonding have been highlighted, and metavalent bonding [33] and, hyperbonding [34] have been used to describe the root cause of these unusually large optical matrix elements. Importantly, delocalized p-orbital bonds in crystalline Ge2Sb2Te5 are strongly disturbed by femtosecond laser pulses [13]. The femtosecond laser pulse changes the optical properties of crystalline GeTe–Sb2Te3 alloys without perturbing the lattice until much later, in the picosecond time range. Moreover, the material can return to its original crystalline optical state when it does not accumulate sufficient heat energy from the laser pulse to amorphize it. This suggests that Sb2Te3–GeTe materials can perform nonlinear optical volatile tunability. The delay in lattice perturbation is also an advantage as it reduces switching variability. This is because the refractive index of the waveguide will vary when subjected to thermal expansion. By delaying lattice perturbation, the optical signal will not be affected by a change in refractive index of the waveguide during the femtosecond laser switching process. Moreover, since Ge2Sb2Te5 does not change structural phase, stress is minimized, and we expect the cyclability endurance to be higher than that of phase change-tuned materials. Therefore, we propose to use femtosecond laser switching of Ge2Sb2Te5 to implement a purely optical NLAF.

Integrating both PCM-based neural network components on the same chip requires the photonic platform to support both functionalities. Thus, optimizing the waveguide material of the chip becomes critical. The waveguide material should exhibit a large bandgap to support visible and near infrared (N-IR) operations. This is because the weighting component must operate in the visible region as the Ge2Sb2Te5 NLAF component requires an 800 nm wavelength femtosecond optical signal for switching [13]. Moreover, the NLAF component is set to operate in the telecommunication wavelength (N-IR region) as it can be easily integrated with other Si or SiN-based photonic components commonly found on a photonic integrated chip (PIC). In addition, operating in the N-IR wavelength is more efficient as Ge2Sb2Te5 exhibits lower optical losses in the near infrared compared to the visible [28]. The waveguide material should also be optimized to efficiently dissipate heat from the Ge2Sb2Te5 material to minimize network latency. Hence, waveguide materials that have a high heat conductivity are necessary.

To choose an appropriate waveguide material for this setup, we compared the properties of four prototypical waveguide materials: Si, SiN, AlN and diamond [3538]. Table 1 shows the optical and thermal properties of the waveguide materials. A Ge2Sb2Te5/waveguide stack shown in the inset of Figure 2 was modeled to understand how the waveguide material affects Ge2Sb2Te5 heat dissipation. In the model, we set the Ge2Sb2Te5 layer to be at 615 °C, which is its melting point, and let the structure cool to room temperature. The temporal variation in temperature of the Ge2Sb2Te5 layer was modeled using the finite element methods (FEM) on COMSOL to solve the heat diffusion equation. More information on the simulation parameters can be found in the Supplementary Material Section 4. From the simulation, we found that the waveguide material could no longer improve the heat dissipation rate of the Ge2Sb2Te5 layer when its thermal conductivity is in the range of 102–103 W/mK. Instead, the heat dissipation rate is mostly limited by the Ge2Sb2Te5 layer itself. This is evident in Figure 2, where the Ge2Sb2Te5/diamond stack was only 32 °C lower than the Ge2Sb2Te5/Si stack at 1 ns whilst the Ge2Sb2Te5/Si stack was 50 °C lower than the Ge2Sb2Te5/SiN stack. Note, the thermal conductivity of diamond is an order of magnitude higher than Si whilst thermal conductivity of Si is only 5 times higher than SiN. Hence the waveguide materials Si, AlN and diamond are suitable to efficiently dissipate heat from the Ge2Sb2Te5 layer. However, Si does not transmit visible light and diamond is not compatible with CMOS processing. Moreover, complex fabrication processes are required to obtain single-crystal diamond waveguides that exhibit the superior thermal properties found in Table 1. Ultimately, we chose AlN as the waveguide material due to its large bandgap, ability to transmit in the visible and N-IR, low thermo-optic coefficient, CMOS compatibility, and its ability to efficiently dissipate heat from the Ge2Sb2Te5 material.

Table 1:

Comparison of optical and thermal properties of waveguide materials.

Waveguide material Refractive index at 1550 nm Transparency wavelength region (µm) CMOS compatibility Thermal conductivity (W/mK) Thermo-optic coefficient (/K)
Si [3941] 3.47 1.2–8 Yes 131 1.8 × 10−4 (@ 1550 nm)
SiN [39, 42, 43] 2.0 0.4–4.6 Yes 20 2.45 × 10−5 (@ 1550 nm)
AlN [35, 39, 44, 45] 2.12 0.2–13.6 Yes 285 2.3 × 10−5 (@ 1560 nm)
Diamond [36, 39, 46, 47] 2.38 0.22–50 No 2000 0.6 × 10−5 (@ 1560 nm)
Figure 2: 
Temperature variation of the Ge2Sb2Te5 film with time as it cools. Figure inset shows the Ge2Sb2Te5/waveguide stack simulated. More information of the simulation parameters can be found in the Supplementary Material Section 4.
Figure 2:

Temperature variation of the Ge2Sb2Te5 film with time as it cools. Figure inset shows the Ge2Sb2Te5/waveguide stack simulated. More information of the simulation parameters can be found in the Supplementary Material Section 4.

3.2 All-optical NLAF base on femtosecond laser switching of Ge2Sb2Te5

A transient change in optical constants is essential for implementing the NLAF. Therefore, we measured the optical properties of crystalline Ge2Sb2Te5 using femtosecond time-resolved pump–probe transmission. Figure 3A shows the pump–probe setup used to switch a crystalline Ge2Sb2Te5 film with a thickness of 27 ± 3 nm. The transient changes to the near-normal reflectivity and transmissivity were simultaneously recorded by a 1500 nm wavelength time-delayed probe with a pump at 800 nm wavelength. The probe was set to 1500 nm wavelength to match the nanophotonic neural network operating wavelength. The dielectric function was then obtained by numerically inverting the Fresnel equations for the multi-layer structure, assuming all changes occurred in the crystalline Ge2Sb2Te5 film. From Figure 3B and C, we observed that the maximum change in the dielectric constants occurred within 1 ps followed by a subsequent relaxation to the original state. The change in optical properties within 1 ps suggests that the material is a good candidate for the NLAF component. Figure 3D and E show the corresponding optical constants at 1 ps for the different laser pump fluence used.

Figure 3: 
Femtosecond time-resolved pump–probe measurement of crystalline Ge2Sb2Te5. (A) Pump–probe experimental setup. Variation in the (B) real and (C) imaginary part of the dielectric function with time at 1500 nm wavelength when Ge2Sb2Te5 is excited with a 35 fs laser pulse at 800 nm wavelength. Corresponding (D) refractive index, n, and (E) extinction coefficient, k, of Ge2Sb2Te5 at 1 ps.
Figure 3:

Femtosecond time-resolved pump–probe measurement of crystalline Ge2Sb2Te5. (A) Pump–probe experimental setup. Variation in the (B) real and (C) imaginary part of the dielectric function with time at 1500 nm wavelength when Ge2Sb2Te5 is excited with a 35 fs laser pulse at 800 nm wavelength. Corresponding (D) refractive index, n, and (E) extinction coefficient, k, of Ge2Sb2Te5 at 1 ps.

At different wavelengths, the interplay between free carrier absorption and the disruption of metavalent bonds/hyperbonds upon laser excitation have different effects on the crystalline Ge2Sb2Te5 optical response. The free carriers generated from interband transitions lead to free-carrier absorption and thus caused the real part of the dielectric permittivity to decrease and the imaginary part to increase [48]. In contrast, the disruption of metavalent bonds/hyperbond caused both the real and imaginary part of the dielectric permittivity to decrease [15, 49]. Hence the free carrier absorption effect is more pronounced at the 1500 nm wavelength as we see a decrease in the real part of the dielectric function while the imaginary part increases. However, at the 800 nm wavelength the effects of metavalent/hyperbond disruption outweigh free carrier absorption as we see both the real and imaginary part of the dielectric constant decrease [13]. Note, the laser pump in both scenarios were from the same wavelength, at 800 nm.

The transient optical constants of the Ge2Sb2Te5 film, see Figure 3D and E, were incorporated into an AlN micro-ring resonator (MRR). Note, although the material below the Ge2Sb2Te5 film is now AlN, the Ge2Sb2Te5 dielectric constant derived in Figure 3D and E could be used in this analysis. This is because we have isolated the substrate from the Ge2Sb2Te5 film when we used Fresnel equations to derive the Ge2Sb2Te5 dielectric constants. The choice of photonic device can influence the resultant NLAF transfer function. Supplementary Material Section 5 discusses this in further details. In brief, a resonating structure was needed to enhance the nonlinearity of the NLAF component. This is because a straight waveguide structure had a dynamic range of less than 1 dB, as shown in Figure S6 of the Supplementary Material. A NLAF with such a small dynamic range would be susceptible to noise and the signal would need to be amplified. This amplification would increase the overall circuit complexity and power consumption as active amplifiers would be added to each NLAF component.

By combining the Ge2Sb2Te5 film with a MRR, a sigmoid transmission response with a dynamic range of 9.7 dB was achieved for the NLAF component. The dynamic range was higher than the sigmoid functions reported in reference [10]. Figure 4A shows the Ge2Sb2Te5-based MRR NLAF. The dimensions of the AlN waveguide and Ge2Sb2Te5 film in Figure 4A were optimized to ensure single mode operation at the 1500 nm wavelength. The ring radius was chosen to be 100 µm [26] and the coupling gap between the ring and waveguide was 140 nm to achieve critical coupling. The Ge2Sb2Te5 film length was 100 nm. The spectral response of the photonic device was modeled using Eqs. (3)(6) to derive the NLAF. Figure 4B and C show the spectrum of the unperturbed device and the device spectral change when excited with various femtosecond laser fluence pulses, respectively. From Figure 4C, the resonance peak of the MRR experienced a blue-shift with increasing laser fluence. This is because the real part of the dielectric function, and thus n, of the Ge2Sb2Te5 layer decreased (Figure 3B), which led to a decrease in n eff , GST found in Eq. (4). This in turn caused the phase shift value, ϕ , in Eq. (4) to decrease thus resulting in a blue-shift in the resonance peak. The blue-shift gave a positive sigmoid function. As the NLAF input is a single wavelength signal, the transfer function of the NLAF will be the change in transmission values for the different laser fluence at a particular wavelength. The sigmoidal NLAF in Figure 4D was obtained by plotting the transmission value of the corresponding laser fluence used at 1499.806 nm wavelength. The black dotted line in Figure 4C intersects the corresponding transmission values of the different laser fluence pulses at the 1499.806 nm wavelength. We chose this wavelength to obtain a positive sigmoid function as the lowest fluence corresponds to the lowest transmission value. Note, the positive sigmoid function is a commonly used NLAF function in machine learning algorithms [50].

Figure 4: 
Ge2Sb2Te5-based MRR NLAF. (A) Schematic diagram of MRR with the corresponding dimensions. Spectral response of the Ge2Sb2Te5-based MRR NLAF when it is (B) unperturbed and (C) excited with various femtosecond laser fluence pulses. (D) Resulting sigmoid function used to implement the NLAF. The transmission values of the blue data points intersect with the black dotted line in (C).
Figure 4:

Ge2Sb2Te5-based MRR NLAF. (A) Schematic diagram of MRR with the corresponding dimensions. Spectral response of the Ge2Sb2Te5-based MRR NLAF when it is (B) unperturbed and (C) excited with various femtosecond laser fluence pulses. (D) Resulting sigmoid function used to implement the NLAF. The transmission values of the blue data points intersect with the black dotted line in (C).

The resonance peak amplitude variation of the MRR upon femtosecond laser excitation had similar trends to a PCM-tuned MRR that underwent permanent structural phase transition from the crystalline to amorphous state [51], [52], [53]. We observed a blue shift in the resonance peak as the laser fluence increased. Moreover, the extinction ratio and the Q-factor of the device generally increased. Note, this increase is smaller than nonvolatile devices as a permanent structural phase transition caused the Ge2Sb2Te5 material to have a larger change in n and k. The calculated Q-factor and extinction ratio of the MRR for the various laser fluence can be found in Table S4. Although the imaginary part of the Ge2Sb2Te5 material dielectric function, and thus k, increased with increasing fluence, the Q-factor and extinction ratio still increased. This is because n of the Ge2Sb2Te5 layer also decreases, and the mode profile shifts downward in the waveguide. The downward shift was evident when we see a decreasing n eff value of the Ge2Sb2Te5-tuned waveguide as the laser fluence increased in Table S2. Moreover, this effect is also illustrated in Figure S3 where we compared the modal profile of the unperturbed MRR and the MRR excited with an 11.4 mJ/cm2 laser pulse. As the mode became less confined to the lossy Ge2Sb2Te5 layer, the imaginary part of the effective refractive index decreased. Hence α GST in Eq. (5) decreases with increasing laser fluence, resulting in a higher Q-factor and extinction ratio. As previously mentioned in Section 2, the corresponding α GST can be found in Table S3. From this result, we observed that the increase in k of the Ge2Sb2Te5 material did not compromise the MRR performance. Instead, the Q-factor and extinction ratio increased, which was similar, but smaller, to the trends observed in nonvolatile Ge2Sb2Te5-tuned MRR devices.

Implementing the NLAF model on a chip to achieve sub-picosecond optical switching requires cutting-edge PIC technology. Femtosecond on-chip lasers are required for an on-chip NLAF. Current femtosecond switching operations are done in free space and most of them require bulky setups. Nevertheless, developing on-chip femtosecond lasers has started to gain traction [54] and it is likely that this model can be implemented on a chip in the future. In the short term, femtosecond fiber lasers [55] could be used as optical signal inputs, which are small enough to be utilized by data centers.

The computation speed of a feedforward neural network depends on the rate that the NLAF can reset, and this in turn depends on the Ge2Sb2Te5 cooling rate. The heat energy from the Ge2Sb2Te5 layer needs to be dissipated before subsequent pulses can be fired to prevent cumulative heating and subsequent melting. Therefore, we must determine the maximum repetition rate that the crystalline Ge2Sb2Te5 film can be excited.

A two-temperature model [32, 56] was used to describe the thermal response of Ge2Sb2Te5 during the femtosecond laser switching process. As a femtosecond laser pulse is much shorter than the duration for which the electron and phonons (lattice) temperature equilibrate, there will be a time delay between the laser excitation process and the increase in lattice temperature. This effect is evident in references [13, 56, 57], where the Ge2Sb2Te5 material reaches its equilibrium temperature at the picosecond time scale when excited by a femtosecond laser pulse. The model accounts for this delay by considering the heat transfer from the electrons to lattice during the laser switching process. More information about the two-temperature model analysis can be found in the Supplementary Material Section 6. To validate the simulation model, we first model the femtosecond laser switching experiment of the Ge2Sb2Te5 thin films described in Figure 3A and compared the temperature with measured data by Waldecker et al. [13]. Note, the pump–probe setup and sample were similar in both experiments. The Ge2Sb2Te5 lattice temperature in this work was found to match the temperature values reported by Waldecker et al. Figure 5A compares the simulated temperature and the measured data trendline. The blue data points in Figure 5A were the average temperature across the film surface that was exposed to the laser beam. The temporal change in Ge2Sb2Te5 temperature was also simulated and shown in Figure 5B. When we fitted the time-dependent temperature equation reported by Waldecker et al., we found that the average heating rate across all the laser fluence used was 2.15 ps, with a standard deviation of 0.01 ps. Moreover, the amplitude of the cooling temperature was found to be between 13 and 35% from the increase in temperature. These results were close to the range of values measured by Waldecker et al., where a heating rate of 2.2 ps and cooling temperature range of 20–50% were reported. Thus, this agreement adds a degree of confidence to our heating simulation and concomitant NLAF model.

Figure 5: 
Two-temperature heat simulation of the crystalline Ge2Sb2Te5 material when excited with femtosecond laser pulses. (A) Average temperature of the Ge2Sb2Te5 film after being exposed to the various laser fluence used in Figure 3. (B) Corresponding variation of Ge2Sb2Te5 temperature with time. Both the temperature values and heat dissipation rates match those reported in reference [13]. (C) Average temperature of the Ge2Sb2Te5 film on AlN MRR waveguide when the NLAF is subjected to 1 GHz femtosecond laser pulse train. (D) Heat distribution cross-section of the Ge2Sb2Te5 NLAF MRR at different times after being excited with a 35 fs laser pulse.
Figure 5:

Two-temperature heat simulation of the crystalline Ge2Sb2Te5 material when excited with femtosecond laser pulses. (A) Average temperature of the Ge2Sb2Te5 film after being exposed to the various laser fluence used in Figure 3. (B) Corresponding variation of Ge2Sb2Te5 temperature with time. Both the temperature values and heat dissipation rates match those reported in reference [13]. (C) Average temperature of the Ge2Sb2Te5 film on AlN MRR waveguide when the NLAF is subjected to 1 GHz femtosecond laser pulse train. (D) Heat distribution cross-section of the Ge2Sb2Te5 NLAF MRR at different times after being excited with a 35 fs laser pulse.

The two-temperature model was then applied to the NLAF AlN ring resonator structure to determine the maximum repetition rate that the crystalline Ge2Sb2Te5 film can be excited without melting. A train of laser pulses were fired when the material reached room temperature. The laser beam spot was set to 1 um on the waveguide setup. In the simulation, laser pulses were set to the maximum fluence (11.4 mJ/cm2) that was needed to program the MRR NLAF. This fluence leads to the slowest operating scenario because the longest cooling time results. Figure 5C shows the average temperature of the Ge2Sb2Te5 film with respect to time when excited with 1 GHz laser pulses at 11.4 mJ/cm2. Figure 5D shows the resulting cross-sectional heat distribution in the photonic NLAF MRR at different times. The heat distribution in Figure 5D has a similar switching behavior to that of the thin film sample. At 1 ps, the electrons transfer heat to the lattice, and we see the Ge2Sb2Te5 film temperature increases. The material reaches its equilibrium temperature at 10 ps, after which the heat dissipates into the surrounding. The 100 ps profile exemplifies how heat dissipates from the Ge2Sb2Te5 film. The device reached room temperature at 1 ns. Thus, the AlN waveguide was able to efficiently dissipate the heat, allowing the NLAF component to reset in 1 ns. The NLAF operating speed can be faster when sequential pulses are fired before the Ge2Sb2Te5 film reaches room temperature. However, the structural properties of Ge2Sb2Te5 may change when subjected to long-term cumulative heating and this in turn could alter the behavior of device. Future works should focus on the cyclability of the material and understand how the optical properties of crystalline Ge2Sb2Te5 changes when held above room temperature for long periods of time.

Other ways to improve the operating speed of the NLAF involves optimizing the PCM or introducing redundancy in the circuit. The PCM can be substituted with a chalcogenide that has a higher thermal conductivity. In Figure 2, we show that the heat dissipation is mainly limited by the FCC Ge2Sb2Te5 alloy when the waveguide material is optimized. Hence, chalcogenides with a higher heat conductivity can be considered. For instance, the in-plane heat conductivity of Sb2Te3 films at room temperature is 5 W/m⋅K [58], which is more than twice that of Ge2Sb2Te5 [59]. However, there is a design conflict as a higher thermal conductivity material would also require higher fluence laser pulses to excite the material. Ultimately, the PCM should be optimized to be within the power budget and efficiently dissipate the heat energy generated from the laser pulse. Another solution involves introducing redundancy in the perceptron. Extra NLAF components can be introduced within the perceptron to accelerate the programming process. For instance, introducing three additional NLAF components to one perceptron can enhance programming speed by four times as there will be at least one NLAF component that can be programmed every 250 ps. However, this requires a larger footprint and may not be ideal in large scale networks.

3.3 Neural network design and performance

An Sb2S3-tuned directional coupler for the MZI weighting component, shown in Figure 6A, was designed to complement the NLAF. The design method can be found in references [27, 28]. Figure 6B and C show the performance of the AlN directional coupler when Sb2S3 is in the amorphous and crystalline states, respectively. The directional coupler displayed low insertion losses (<−1.5 dB) and cross talk (−15 dB to −30 dB) for both amorphous and crystalline states in the 770–830 nm wavelength range. Multi-bit switching was introduced by amorphizing 31 sections of the 24.896 µm-long PCM strip. Each section was amorphized at 0.8 µm intervals. Note, the directional coupler has 32 switching states (5 bits) when the purely crystalline state is included. In this work, the focused laser beam, which is used to amorphise the Sb2S3 sections, is assumed to have a 0.8 µm spot size. However, the strip can also be programmed using a microheater [52, 53, 60, 61]. Microheaters can be programmed to tune the growth of the crystal Sb2S3 because the crystallization process is growth dominated [28]. Nonplasmonic materials like W make suitable microheater materials as they show minimal insertion losses [62, 63]. Computations done in reference [63] revealed that W microheaters only had an additional insertion loss of 0.1 dB/μm in the TM mode. Note, the position of the W electrodes were also optimized to minimize modal interference. Figure 6D shows the transmission values of the bar and cross ports for different amorphized lengths at the 800 nm wavelength. The corresponding power field profile of the device for the five states labeled as (i)–(v) in Figure 6D, are shown in Figure 6E.

Figure 6: 
1X2 Sb2S3-tuned directional coupler for the MZI weights. (A) Schematic diagram of the directional coupler device. Performance of the directional coupler across 770–830 nm wavelength in the (B) amorphous and (C) crystalline state. (D) Transmission values at the different ports for different amorphized Sb2S3 length at the 800 nm wavelength. (E) Corresponding power field profile of the device for the five states labeled in (D).
Figure 6:

1X2 Sb2S3-tuned directional coupler for the MZI weights. (A) Schematic diagram of the directional coupler device. Performance of the directional coupler across 770–830 nm wavelength in the (B) amorphous and (C) crystalline state. (D) Transmission values at the different ports for different amorphized Sb2S3 length at the 800 nm wavelength. (E) Corresponding power field profile of the device for the five states labeled in (D).

Implementing a multi-layer neural network involves cascading the perceptrons, where each perceptron constitutes the PCM-tuned weights and NLAF. The overall schematic of a three-layer neural network is shown in Figure 7A. Figure 7B shows the neural network components in each neural layer. The colored circles in Figure 7B correspond to the colors used to represent the different neural layers in Figure 7A. Starting from the perceptrons in the first neural layer (red dots), the 800 nm output of the weighting component programs the Ge2Sb2Te5 layer of the NLAF. The 1500 nm output signal from the NLAF then propagates into the second 100 by 100 weighting layer, which is represented by the green dots. To avoid upconversion from 1500 to 800 nm, the weights in the second layer will operate at the 1500 nm wavelength range. The design for the weighting components in the C-band can be found in reference [28]. Upon implementing the weighting computations, the optical signal is then converted back to the electrical domain to implement the rest of the network digitally. Note, the optical signal has to be converted back to the electrical domain to implement the softmax function. Implementing it before the second NLAF layer is ideal as the upconversion of laser signal, which consumes a large amount of power [64], will be avoided. Additional neural layers can be cascaded before the proposed three - layer neural network shown in Figure 7A. The NLAF components in these additional layers will be set to operate at the 800 nm wavelength. This allows the NLAF output signal to propagate directly to the next weighting connection. Moreover, this configuration allows a new set of signals to be generated at each neural layer, specifically at the NLAF input port. The transient optical switching response of Ge2Sb2Te5 measured at 800 nm pump and probe pulses can be found in reference [13]. In this configuration, the penultimate NLAF layer is also set to 1500 nm wavelength and the last NLAF layer is done digitally. Having the penultimate NLAF layer operating at the 1500 nm wavelength is critical as the signal will be integrated with other photonic components involved in the optical–electrical conversion.

Figure 7: 
Chalcogenide-tuned neural network used to run MNIST dataset. (A) Schematic of the neural network. (B) Neural network components in each neural layer. The colored circles represent the corresponding neural layers found in (A). (C) Corresponding network accuracy for the different switching errors in the weights and NLAF.
Figure 7:

Chalcogenide-tuned neural network used to run MNIST dataset. (A) Schematic of the neural network. (B) Neural network components in each neural layer. The colored circles represent the corresponding neural layers found in (A). (C) Corresponding network accuracy for the different switching errors in the weights and NLAF.

The chalcogenide-tuned neural network shown in Figure 7A was modeled to demonstrate its performance. The transfer functions of the components were obtained through curve fitting the data points in Figure 4D and 6D. The NLAF transmission, T NLAF , as a function of fluence, F, is given by Eq. (7). The weights (P3A) transmission for the bar, T bar , and cross, T cross , ports as a function of the amorphized Sb2S3 length, x, along the coupler are given in Eqs. (8) and (9), respectively. As can be seen, sigmoid and polynomial functions gave the best least-squares fitting results.

(7) T NLAF = 0.7478 1 + 2.728 ( e 0.7522 ( F 2.4486 ) ) + 0.04352

(8) T bar = 0.8207 1 + 0.0139 ( e 0.09293 ( x + 0.9209 ) )

(9) T cross = 2.845 × 10 9 x 4 2.92 × 10 7 x 3 + 0.0002054 x 2 0.005326 x + 0.02582

A three-layer fully connected neural network consisting of all-chalcogenide perceptrons was trained on Google TensorFlow with the standard MNIST dataset, which contained 60,000 grayscale images of handwritten digits. The first layer of the network consisted of 100 neurons to receive the image pixels. Similarly, the second layer composed of 100 neurons. Each neuron was connected to all the neurons from the first layer, giving a 100 by 100 connection matrix. The third layer contained just 10 neurons to represent the outputs 0 to 9. The all-optical NLAF (second box in Figure 7B) was connected to the weighting connections found in the first two consecutive neural layers [65].

To test the robustness of this network, we monitor the inference accuracy as a function of switching variability (i.e., noise) from both the P3A and the NLAF [66]. Figure 7C shows the corresponding network accuracy for the different noise generated from the P3A and NLAF. Unlike digital electronics, where noise is artificially added to the training data to increase network robustness, analogue photonic neural networks have intrinsic noise. The highest inference accuracies were achieved when the standard deviation values were 0.5 and 1.5 for the weights and NLAF, respectively. This corresponds to a Signal to Noise Ratio (SNR) of 6660 for the NLAF and 20,000 for the weights. Hence, it may be possible to deliberately add training noise to fine-tune the network for inference operations on physical noisy input signals. However, we found that when the training noise SNR was less than 2200 for both components (standard deviation values exceeding 4.5), the inference accuracy starts to decrease indefinitely.

The all-chalcogenide all-optical perceptron offers an energy-efficient and fast neural network by avoiding optical to electrical to optical conversions. Only at the penultimate neural layer, the optical signal is converted to the electrical domain to implement the softmax function. The signal remains in the electrical domain after the conversion and further electro-optic conversions are not required. This reduces network latency and energy consumption. Supplementary Material Section 7 and Table S6 compare the network performance with current state of the art electro-optical and all-optical perceptron. The slowest operation time possible of each NLAF is 1 ns, assuming that time is required for the Ge2Sb2Te5 to cool to room temperature before the next pulse is fired, and that all NLAF use the maximum programming fluence of 11.4 mJ/cm2. In practice, the NLAF operating speed could be faster as not all NLAFs will need a programming laser fluence of 11.4 mJ/cm2. Hence the Ge2Sb2Te5 temperature will not reach its maximum and a shorter time will be required to cool. Moreover, the operating speed can be further increased when subsequent pulses are fired without waiting for the material to cool to room temperature. Since the weights consist of passive devices, the overall network energy is mainly determined by the optical NLAF. This constitutes the femtosecond pulse generation to program the Ge2Sb2Te5 layer and a static input laser signal for the MRR. Note, the NLAF programming pulse is not electrically modulated, and it comes directly from the weighting output. Amplification of the signal may be required but this can be done entirely in the optical domain. For the optical NLAF programming pulse, the maximum energy consumed for the whole network is only 0.90 µJ assuming a total of 10,000 NLAF components, each using the maximum programming fluence of 11.4 mJ/cm2 focused to a Gaussian with a beam diameter of 1 µm. The static input signal for the NLAF MRR can be implemented with on-chip lasers, which typically have a high Wall Plug Efficiency (WPE) [67]. For instance, a III-IV based Si on-chip C-band laser can have a WPE of 20% [68, 69]. Assuming that the optical input signal of a NLAF MRR is 1 mW [70], [71], a 100 by 100 weighting connection layer that corresponds to 10,000 NLAF components would require an electrical input of 50 W to power the lasers. This is substantially lower than von Neumann technologies like a GPU, which has a power consumption of up to 300 W [72]. The Sb2S3-tuned MZI is nonvolatile and can preserve the optical weights. Thus, no additional power is required when performing the weighting function under the feedforward propagation mode.

4 Conclusion

To conclude, by combining AlN waveguides with two different effects in chalcogenides, we show that a passive, efficient, and accurate all-optical perceptron is feasible. For the first time, we demonstrate that the disruption of delocalized p-orbital metavelent bonds with a femtosecond laser can be used to implement a photonic NLAF device with sub-picosecond dielectric permittivity switching response. Moreover, the low–loss Sb2S3 material used to set the network interconnection MZI weights gave low insertion loss and cross talk. The proposed all-optical perceptron model can also be extended to the terahertz frequency in the future, as a recent study demonstrated volatile and nonvolatile switching of Ge2Sb2Te5 [73]. In the feedforward mode, the heat dissipation of the NLAF material is the bottleneck of the network. To improve the NLAF efficiency, future works should study the cyclability of Ge2Sb2Te5 material switching in the femtosecond time range and how the thermal properties of the material changes when it accumulates heat energy. The speed of the NLAF can also be improved by considering other PCMs, like Sb2Te3, which has a higher thermal conductivity, or by optimizing the architecture of the network to accommodate multiple NLAF components. Ultimately, this depends on the design requirements, particularly the network power budget and chip footprint.

Corresponding authors: Ting Yu Teo and Robert E. Simpson, Singapore University of Technology and Design, 8 Somapah Road, Singapore 487372, Singapore, E-mail: , ; and Volker J. Sorger, Deptartment of Electrical and Computer Engineering, George Washington University, Washington, DC, USA, E-mail:
Ting Yu Teo and Xiaoxuan Ma contributed equally to this work.

Funding source: Agencia Española de Investigación

Award Identifier / Grant number: IJC2018-037384-I

Award Identifier / Grant number: IJC2018-037384-I

Award Identifier / Grant number: FA9550-20-1-0193

Award Identifier / Grant number: #A18A7b0058

  1. Author contributions: All the authors have accepted responsibility for the entire content of this submitted manuscript and approved submission.

  2. Research funding: The NLAF design, Sb2S3 MZI weights, material growth, and thermal modeling were supported by the Agency for Science, Technology and Research (A*STAR) under the Advanced Manufacturing and Engineering (AME) grant #A18A7b0058. The network training was supported from the Presidential Early Career Award for Scientist and Engineers (PECASE) nominated by the Department of Defense through the Air Force Office of Scientific Research under award number FA9550-20-1-0193. The pump–probe measurements were funded by Fundació Cellex, Fundació Mir-Puig, and Generalitat de Catalunya through CERCA. E.P acknowledges the support from IJC2018-037384-I funded by MCIN/AEI/10.13039/501100011033.

  3. Conflict of interest statement: The authors declare no conflicts of interest regarding this article.


[1] D. S. Jeong, K. M. Kim, S. Kim, B. J. Choi, and C. S. Hwang, “Memristors for energy-efficient new computing paradigms,” Adv. Electron. Mater., vol. 2, no. 9, p. 1600090, 2016. in Google Scholar

[2] Y. Shen, N. C. Harris, S. Skirlo, et al.., “Deep learning with coherent nanophotonic circuits,” Nat. Photonics, vol. 11, no. 7, pp. 441–446, 2017. in Google Scholar

[3] J. Feldmann, N. Youngblood, M. Karpov, et al.., “Parallel convolutional processing using an integrated photonic tensor core,” Nature, vol. 589, no. 7840, pp. 52–58, 2021. in Google Scholar PubMed

[4] H. Zhang, M. Gu, X. D. Jiang, et al.., “An optical neural chip for implementing complex-valued neural network,” Nat. Commun., vol. 12, no. 1, p. 457, 2021. in Google Scholar PubMed PubMed Central

[5] M. Miscuglio, G. C. Adam, D. Kuzum, and V. J. Sorger, “Roadmap on material-function mapping for photonic-electronic hybrid neural networks,” Apl. Mater., vol. 7, no. 10, p. 100903, 2019. in Google Scholar

[6] Y. Zuo, B. Li, Y. Zhao, et al.., “All-optical neural network with nonlinear activation functions,” Optica, vol. 6, no. 9, pp. 1132–1137, 2019. in Google Scholar

[7] I. A. D. Williamson, T. W. Hughes, M. Minkov, B. Bartlett, S. Pai, and S. Fan, “Reprogrammable electro-optic nonlinear activation functions for optical neural networks,” IEEE J. Sel. Top. Quant. Electron., vol. 26, no. 1, pp. 1–12, 2020. in Google Scholar

[8] A. Jha, C. Huang, and P. R. Prucnal, “Reconfigurable all-optical nonlinear activation functions for neuromorphic photonics,” Opt. Lett., vol. 45, no. 17, pp. 4819–4822, 2020. in Google Scholar

[9] G. Mourgias-Alexandris, A. Tsakyridis, N. Passalis, A. Tefas, K. Vyrsokinos, and N. Pleros, “An all-optical neuron with sigmoid activation function,” Opt Express, vol. 27, no. 7, pp. 9620–9630, 2019. in Google Scholar

[10] M. Miscuglio, A. Mehrabian, Z. Hu, et al.., “All-optical nonlinear activation function for photonic neural networks [Invited],” Opt. Mater. Express, vol. 8, no. 12, pp. 3851–3863, 2018. in Google Scholar

[11] C. Ríos, N. Youngblood, Z. Cheng, et al.., “In-memory computing on a photonic platform,” Sci. Adv., vol. 5, no. 2, p. eaau5759, 2019. in Google Scholar PubMed PubMed Central

[12] K. Tanaka, “Optical nonlinearity in photonic glasses,” in Springer Handbook of Electronic and Photonic Materials, New York, Springer, 2007, p. 1063.10.1007/978-0-387-29185-7_44Search in Google Scholar

[13] L. Waldecker, T. A. Miller, M. Rude, et al.., “Time-domain separation of optical properties from structural transitions in resonantly bonded materials,” Nat. Mater., vol. 14, no. 10, pp. 991–995, 2015. in Google Scholar PubMed

[14] N. Youngblood, C. Ríos, E. Gemo, et al.., “Tunable volatility of Ge2Sb2Te5 in integrated photonics,” Adv. Funct. Mater., vol. 29, no. 11, p. 1807571, 2019. in Google Scholar

[15] K. Shportko, S. Kremers, M. Woda, D. Lencer, J. Robertson, and M. Wuttig, “Resonant bonding in crystalline phase-change materials,” Nat. Mater., vol. 7, p. 653, 2008. in Google Scholar PubMed

[16] A. V. Kolobov, M. Krbal, P. Fons, J. Tominaga, and T. Uruga, “Distortion-triggered loss of long-range order in solids with bonding energy hierarchy,” Nat. Chem., vol. 3, no. 4, pp. 311–316, 2011. in Google Scholar PubMed

[17] R. E. Simpson, P. Fons, X. Wang, A. V. Kolobov, T. Fukaya, and J. Tominaga, “Non-melting super-resolution near-field apertures in Sb–Te alloys,” Appl. Phys. Lett., vol. 97, no. 16, p. 161906, 2010. in Google Scholar

[18] J. K. Behera, W. Wang, X. Zhou, et al.., “Resistance modulation in Ge2Sb2Te5,” J. Mater. Sci. Technol., vol. 50, pp. 171–177, 2020. in Google Scholar

[19] N. Yamada, E. Ohno, K. Nishiuchi, N. Akahira, and M. Takao, “Rapid-phase transitions of GeTe-Sb2Te3 pseudobinary amorphous thin films for an optical disk memory,” J. Appl. Phys., vol. 69, no. 5, pp. 2849–2856, 1991. in Google Scholar

[20] M. Born and E. Wolf, Principles of Optics, Cambridge, Cambridge University Press, 2006.Search in Google Scholar

[21] W. K. Njoroge, H.-W. Wöltgens, and M. Wuttig, “Density changes upon crystallization of Ge2Sb2.04Te4.74 films,” J. Vac. Sci. Technol., vol. 20, no. 1, pp. 230–233, 2002. in Google Scholar

[22] W. Bogaerts, P. De Heyn, T. Van Vaerenbergh, et al.., “Silicon microring resonators,” Laser Photon. Rev., vol. 6, no. 1, pp. 47–73, 2012. in Google Scholar

[23] J. Pello, Building Up a Membrane Photonics Platform in Indium Phosphide, Eindhoven, Technische Universiteit Eindhoven, 2014.Search in Google Scholar

[24] S. Adachi, “Aluminum nitride (AIN),” in Optical Constants of Crystalline and Amorphous Semiconductors: Numerical Data and Graphical Information, S. Adachi, Ed., Boston, MA, Springer US, 1999, pp. 143–151.10.1007/978-1-4615-5247-5_15Search in Google Scholar

[25] H. R. Philipp, “Silicon dioxide (SiO2) (glass),” in Handbook of Optical Constants of Solids, E. D. Palik, Ed., Boston, Academic Press, 1985, pp. 749-763.10.1016/B978-0-08-054721-3.50040-XSearch in Google Scholar

[26] W. H. P. Pernice, C. Xiong, and H. X. Tang, “High Q micro-ring resonators fabricated from polycrystalline aluminum nitride films for near infrared and visible photonics,” Opt Express, vol. 20, no. 11, pp. 12261–12269, 2012. in Google Scholar PubMed

[27] P. Xu, J. Zheng, J. K. Doylend, and A. Majumdar, “Low-loss and broadband nonvolatile phase-change directional coupler switches,” ACS Photonics, vol. 6, no. 2, pp. 553–557, 2019. in Google Scholar

[28] T. Y. Teo, M. Krbal, J. Mistrik, J. Prikryl, L. Lu, and R. E. Simpson, “Comparison and analysis of phase change materials-based reconfigurable silicon photonic directional couplers,” Opt. Mater. Express, vol. 12, no. 2, pp. 606–621, 2022. in Google Scholar

[29] M. Delaney, I. Zeimpekis, D. Lawson, D. W. Hewak, and O. L. Muskens, “A new family of ultralow loss reversible phase-change materials for photonic integrated circuits: Sb2S3 and Sb2Se3,” Adv. Funct. Mater., vol. 30, no. 36, p. 2002447, 2020. in Google Scholar

[30] W. Dong, H. Liu, J. K. Behera, et al.., “Wide bandgap phase change material tuned visible photonics,” Adv. Funct. Mater., vol. 29, no. 6, p. 1806181, 2019. in Google Scholar

[31] R. E. Simpson and T. Cao, “Phase change material photonics,” in The World Scientific Reference of Amorphous Materials, Singapore, World Scientific, 2020, pp. 487–517.10.1142/9789811215575_0015Search in Google Scholar

[32] H. Liu, W. Dong, H. Wang, et al.., “Rewritable color nanoprints in antimony trisulfide films,” Sci. Adv., vol. 6, no. 51, p. eabb7171, 2020. in Google Scholar PubMed PubMed Central

[33] B. J. Kooi and M. Wuttig, “Chalcogenides by design: functionality through metavalent bonding and confinement,” Adv. Mater., vol. 32, no. 21, p. 1908302, 2020. in Google Scholar PubMed

[34] T. H. Lee and S. R. Elliott, “Chemical bonding in chalcogenides: the concept of multicenter hyperbonding,” Adv. Mater., vol. 32, no. 28, p. 2000340, 2020. in Google Scholar PubMed

[35] N. Li, C. P. Ho, S. Zhu, Y. H. Fu, Y. Zhu, and L. Y. T. Lee, “Aluminium nitride integrated photonics: a review,” Nanophotonics, vol. 10, no. 9, pp. 2347–2387, 2021. in Google Scholar

[36] I. Aharonovich, A. D. Greentree, and S. Prawer, “Diamond photonics,” Nat. Photonics, vol. 5, no. 7, pp. 397–405, 2011. in Google Scholar

[37] D. J. Blumenthal, R. Heideman, D. Geuzebroek, A. Leinse, and C. Roeloffzen, “Silicon nitride in silicon photonics,” Proc. IEEE, vol. 106, no. 12, pp. 2209–2231, 2018.10.1109/JPROC.2018.2861576Search in Google Scholar

[38] R. Soref, “The past, present, and future of silicon photonics,” IEEE J. Sel. Top. Quant. Electron., vol. 12, no. 6, pp. 1678–1687, 2006. in Google Scholar

[39] A. L. Gaeta, M. Lipson, and T. J. Kippenberg, “Photonic-chip-based frequency combs,” Nat. Photonics, vol. 13, no. 3, pp. 158–169, 2019. in Google Scholar

[40] J. Komma, C. Schwarz, G. Hofmann, D. Heinert, and R. Nawrodt, “Thermo-optic coefficient of silicon at 1550 nm and cryogenic temperatures,” Appl. Phys. Lett., vol. 101, no. 4, 2012, Art no. 041905. in Google Scholar

[41] P. Zhou, R. Chen, N. Wang, H. San, and X. Chen, “Reliability design and electro-thermal-optical simulation of bridge-style infrared thermal emitters,” Micromachines, vol. 7, p. 166, 2016. in Google Scholar PubMed PubMed Central

[42] A. Salem, J. Manderscheid, M. Freedman, and J. Gyekenyesi, Reliability Analysis of a Structural Ceramic Combustion Chamber, New York, The American Society of Mechanical Engineers, 1991.10.1115/91-GT-155Search in Google Scholar

[43] A. Arbabi and L. L. Goddard, “Measurements of the refractive indices and thermo-optic coefficients of Si3N4 and SiOx using microring resonances,” Opt. Lett., vol. 38, no. 19, pp. 3878–3881, 2013. in Google Scholar

[44] N. Li, C. P. Ho, Y. Cao, et al.., “Aluminum nitride photonics platforms on silicon substrate,” in OSA Technical Digest, Conference on Lasers and Electro-Optics STh2H.3, Q. Kang, and C. Saraceno, Eds., 2021.10.1364/CLEO_SI.2021.STh2H.3Search in Google Scholar

[45] N. Watanabe, T. Kimoto and J. Suda, “Thermo-optic coefficients of 4H-SiC, GaN, and AlN for ultraviolet to infrared regions up to 500 °C,” Jpn. J. Appl. Phys., vol. 51, 2012, Art no. 112101.10.1143/JJAP.51.112101Search in Google Scholar

[46] Á. I. López-Lorente, M. Karlsson, L. Österlund, and B. Mizaikoff, “Diamond waveguides for infrared spectroscopy and sensing,” in Carbon-Based Nanosensor Technology, C. Kranz, Ed., Cham, Springer International Publishing, 2019, pp. 87–117.10.1007/5346_2017_15Search in Google Scholar

[47] V. Y. Yurov, E. Bushuev, A. Popovich, A. Bolshakov, E. Ashkinazi, and V. Ralchenko, “Near-infrared refractive index of synthetic single crystal and polycrystalline diamonds at high temperatures,” J. Appl. Phys., vol. 122, p. 243106, 2017. in Google Scholar

[48] L. Huang, J. P. Callan, E. N. Glezer, and E. Mazur, “GaAs under intense ultrafast excitation: response of the dielectric function,” Phys. Rev. Lett., vol. 80, no. 1, pp. 185–188, 1998. in Google Scholar

[49] L. Guarneri, S. Jakobs, A. von Hoegen, et al.., “Metavalent bonding in crystalline solids: how does it collapse?” Adv. Mater., vol. 33, no. 39, p. 2102356, 2021. in Google Scholar PubMed

[50] Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, no. 7553, pp. 436–444, 2015. in Google Scholar PubMed

[51] M. Rudé, J. Pello, R. E. Simpson, et al.., “Optical switching at 1.55 μm in silicon racetrack resonators using phase change materials,” Appl. Phys. Lett., vol. 103, no. 14, p. 141119, 2013. in Google Scholar

[52] C. Ríos, Q. Du, Y. Zhang et al.., Ultra-Compact Nonvolatile Photonics Based on Electrically Reprogrammable Transparent Phase Change Materials, arXiv:2105.06010, 2021.10.1186/s43074-022-00070-4Search in Google Scholar

[53] Z. Fang, J. Zheng, A. Saxena, J. Whitehead, Y. Chen, and A. Majumdar, “Non-volatile reconfigurable integrated photonics enabled by broadband low-loss phase change material,” Adv. Opt. Mater., vol. 9, no. 9, p. 2002049, 2021. in Google Scholar

[54] Femtosecond Laser on a Chip, European Commission, 2021. Available at: [accessed: Feb. 6, 2022].Search in Google Scholar

[55] J. Prawiharjo, N. K. Daga, R. Geng, et al.., “High fidelity femtosecond pulses from an ultrafast fiber laser system via adaptive amplitude and phase pre-shaping,” Opt Express, vol. 16, no. 19, pp. 15074–15089, 2008. in Google Scholar PubMed

[56] X. Sun, M. Ehrhardt, A. Lotnyk, et al.., “Crystallization of Ge2Sb2Te5 thin films by nano- and femtosecond single laser pulse irradiation,” Sci. Rep., vol. 6, p. 28246, 2016. in Google Scholar PubMed PubMed Central

[57] Y. H. Wang, F. R. Liu, W. Q. Li, et al.., “Study of non-equilibrium thermal transport in Ge2Sb2Te5 thin films under ultrafast laser excitation using a photo-excited carrier integrated semiconductor model,” J. Appl. Phys., vol. 122, no. 4, 2017, Art no. 043104. in Google Scholar

[58] F. Rieger, K. Kaiser, G. Bendt, et al.., “Low intrinsic c-axis thermal conductivity in PVD grown epitaxial Sb2Te3 films,” J. Appl. Phys., vol. 123, no. 17, p. 175108, 2018. in Google Scholar

[59] M. Kuwahara, O. Suzuki, Y. Yamakawa, et al.., “Temperature dependence of the thermal properties of optical memory materials,” Jpn. J. Appl. Phys., vol. 46, no. 6B, pp. 3909–3911, 2007. in Google Scholar

[60] R. Chen, Z. Fang, J. E. Fröch, P. Xu, J. Zheng, and A. Majumdar, Broadband Nonvolatile Electrically Programmable Silicon Photonic Switches, arXiv e-prints arXiv:2201.05439, 2022.10.1021/acsphotonics.2c00452Search in Google Scholar

[61] J. Zheng, Z. Fang, C. Wu, et al.., “Nonvolatile electrically reconfigurable integrated photonic switch enabled by a silicon PIN diode heater,” Adv. Mater., vol. 32, no. 31, p. 2001218, 2020. in Google Scholar PubMed

[62] M. Miscuglio and V. J. Sorger, “Photonic tensor cores for machine learning,” Appl. Phys. Rev., vol. 7, no. 3, 2020, Art no. 031404. in Google Scholar

[63] M. Miscuglio, J. Meng, O. Yesiliurt, et al.., “Artificial synapse with mnemonic functionality using GSST-based photonic integrated memory,” in 2020 International Applied Computational Electromagnetics Society Symposium (ACES), 2020, pp. 1–3.10.23919/ACES49320.2020.9196183Search in Google Scholar

[64] W. Tan, X. Qiu, G. Zhao, et al.., “High-efficiency frequency upconversion of 1.5 μm laser based on a doubly resonant external ring cavity with a low finesse for signal field,” Appl. Phys. B, vol. 123, no. 2, p. 52, 2017. in Google Scholar

[65] R. Amin, J. K. George, S. Sun, et al.., “ITO-based electro-absorption modulator for photonic neural activation function,” Apl. Mater., vol. 7, no. 8, 2019, Art no. 081112. in Google Scholar

[66] A. Mehrabian, M. Miscuglio, Y. Alkabani, V. J. Sorger, and T. El-Ghazawi, “A winograd-based integrated photonics accelerator for convolutional neural networks,” IEEE J. Sel. Top. Quant. Electron., vol. 26, no. 1, pp. 1–12, 2020. in Google Scholar

[67] Z. Zhou, B. Yin, and J. Michel, “On-chip light sources for silicon photonics,” Light Sci. Appl., vol. 4, no. 11, pp. e358, 2015. in Google Scholar

[68] U. Yutaka, U. Tatsuya, F. Junichi, et al., “High-density optical interconnects by using silicon photonics,” Proc. SPIE, vol. 9010, pp. 901006-1-901006-9, 2014. in Google Scholar

[69] T. Shimizu, N. Hatori, M. Okano, et al.., “High density hybrid integrated light source with a laser diode array on a silicon optical waveguide platform for inter-chip optical interconnection,” in 8th IEEE International Conference on Group IV Photonics, 2011, pp. 181–183.10.1109/GROUP4.2011.6053756Search in Google Scholar

[70] M. Ji, H. Cai, L. Deng, et al., “Enhanced parametric frequency conversion in a compact silicon-graphene microring resonator,” Optics Express, vol. 23, no. 14, pp. 18679–18685, 2015. in Google Scholar PubMed

[71] A. C. Turner, M. A. Foster, A. L. Gaeta, et al., “Ultra-low power parametric frequency conversion in a silicon microring resonator,” Optics Express, vol. 16, no. 7, pp. 4881–4887, 2008. in Google Scholar PubMed

[72] C. Collange, D. Defour, and A. Tisserand, “Power consumption of GPUs from a software perspective,” in Computational Science – ICCS 2009, G. Allen, J. Nabrzyski, E. Seidel, G. D. van Albada, J. Dongarra, and P. M. A. Sloot, Eds., Berlin Heidelberg, Springer, 2009, pp. 914–923.10.1007/978-3-642-01970-8_92Search in Google Scholar

[73] H. Zhu, J. Li, X. Lu, et al.., “Volatile and nonvolatile switching of phase change material Ge2Sb2Te5 revealed by time-resolved terahertz spectroscopy,” J. Phys. Chem. Lett., vol. 13, no. 3, pp. 947–953, 2022. in Google Scholar PubMed

Supplementary Material

The online version of this article offers supplementary material (

Received: 2022-02-22
Revised: 2022-04-02
Accepted: 2022-04-18
Published Online: 2022-05-25

© 2022 Ting Yu Teo et al., published by De Gruyter, Berlin/Boston

This work is licensed under the Creative Commons Attribution 4.0 International License.

Downloaded on 7.12.2023 from
Scroll to top button