1. Introduction

The detection of the relative timing of two signals is commonly required in a wide range of optical measurements and devices, such as in light detection and ranging (LIDAR)-based measurements [1] and clock recovery [2] in optical communication systems. Optical time division multiplexing requires precise relative timing to achieve interleaving accuracy in optical data streams [3]. In addition, good signal synchronization must be ensured to utilize the full bandwidth of multiple channels in optical interconnects. In current optical data communications, jitter detection and clock skew recovery is performed mostly by employing electronic devices with a bandwidth limited to ca. 100 GHz [4], which imposes a bottleneck for ultrahigh speed optical communication systems. To overcome the limitation, optical pulse discriminators have been developed to enable accurate and stable pulse timing manipulation.

Neuromorphic computing is a growing field of research that focuses on mimicking the various neural algorithms, which underlie biological neural networks, by means of artificial synapses. A synapse is junction between two communicating neurons, one of which provides information coded into electrical spike trains (presynaptic neuron) and the other receives them (postsynaptic neuron). The biological synapse is chemical in nature and determines the transmission of spike signal in nonvolatile fashion. This synaptic weight plasticity forms the basis of adaptability, learning, and memory in neural systems. The synaptic weights adjust the strength of connections between neurons in response to spiking activity. Spike-timing-dependent plasticity (STDP) is a mechanism for the implementation of the Hebb’s learning rule, in which synaptic weights are modified according to the temporal correlations between the spikes of pre- and postsynaptic neurons [5]. To date, STDP dynamics have been artificially emulated by the nonlinear properties of electronic devices with an aim toward artificial neuromorphic engineering [6]. All-optical implementation of the STDP mechanism will allow high-capacity signal processing and ultrafast computing device with extended functionality.

An optical pulse timing discriminator has been realized by employing a semiconductor optical amplifier (SOA) [7–10], which provides an output voltage proportional to the relative delay of two pulses. The operation principle relies on the cross-gain modulation, of which the magnitude is dependent on the relative timing of the pulses due to the gain recovery time. Optical implementation of the STDP characteristics using a SOA device has been experimentally demonstrated [11]. Another method utilizes an optical fiber Kerr shutter, which is based on pump-probe nonlinear birefringence [12].

Motivated by the reduction of energy consumption in emulation based on semiconductor optical devices and also by the exploration of a mass-manufacturable platform, alternative architecture for efficient neuromorphic components is under investigation. Amorphous-crystalline phase change in chalcogenides [13–15] and metal-insulator transition in oxides [16] are effectively utilized to implement neuron-inspired functionality due to the large contrast in their physical characteristics and plasticity. These materials also involve threshold behavior in their phase change processes (i.e. the melting point for amorphization and the glass transition temperature for crystallization, and the incubation time required for the production of critical nuclei [17, 18]), which is essential to mimic the threshold firing of neurons. Nonthermal amorphization of chalcogenides [19–23], in which principal structural change is completed at the sub-picosecond level after femtosecond pulse excitation, is advantageous to develop a pulse timing discriminator with a high temporal resolution.

Here, we propose and demonstrate a material-based optical pulse timing discriminator [24] in contrast to the device-based on the use of a SOA. Double thin layers of Ge₂Sb₂Te₅ (GST) and GeCu₂Te₃ (GCT) chalcogenides are employed to discriminate the pulse delay with sub-picosecond time resolution. GCT exhibits an increase of reflectivity upon amorphization [25, 26], whereas the reflectivity is reduced in most of the other Te-based chalcogenides. Therefore, by taking advantage of GeCuTe, the degree of amorphization in the GST/GCT layers is dependent on the arrival order and time delay of two counter-propagating pulses incident from the respective GST and GCT sides, as illustrated in Fig. 1.

 

Fig. 1 Principle for encoding the arrival order of two counter-propagating pulses using a GST/GCT double layer structure.

Download Full Size | PDF

The principle of operation relies on the fact that the amount of absorption of light by the GCT layer changes depending on the phase of underlying GST layer due to the large contrast of refractive index between the crystalline and amorphous phases of GST. Here we define ΔF as the difference in the amount of light absorption by the GCT layer for the GST layer in the crystalline phase and in the amorphous phase. We determine the GST and GCT layer thicknesses such as to maximize ΔF and set the pulse fluence such that the GCT layer is amorphized if the GST layer is crystalline, and the GCT layer remains crystalline (or is partially amorphized if ΔF is not sufficiently large) if the GST layer is amorphous.

2. Amorphization of GeSbTe and GeCuTe

We started with the amorphization of a single layer of GST and a single layer of GCT. 20 nm thick GST and GCT films were sputter-deposited on glass substrates and covered with a 5 nm SiO₂ layer. Both the GST and GCT samples were annealed at 250 °C to obtain a crystalline phase. The photoexcitation source for amorphization was a mode-locked Ti:Sapphire laser operated at a central wavelength of 800 nm with a pulse duration of 150 fs and a repetition rate of 80 MHz. A Pockels cell system was used to extract a single pulse. To write an amorphous mark, the laser beam was focused onto the sample surface using a microscope objective with a numerical aperture of 0.15. A typical fluence for amorphization was 30 mJ/cm² for both the GST and GCT films. A confocal laser scanning microscope (CLSM) at a wavelength of 408 nm was employed for microscopic observation of the amorphous mark.

Figures 2(a) and 2(b) show reflection images of amorphous marks created in the crystalline GST and GCT films by a single femtosecond pulse excitation, respectively. It was confirmed that amorphization was accompanied by a reduction in reflectivity for the GST layer and an increase in reflectivity for the GCT layer [25, 26].

 

Fig. 2 Schematic diagrams of single-layered (a) GST and (b) GCT samples. Scanning laser micrographs of amorphous marks created in (c) GST and (d) GCT layers by single femtosecond pulse excitation.

Download Full Size | PDF

3. Theoretical simulation

Theoretical calculations for a GST/GCT double layer structure were performed to estimate the feasibility of optical pulse discrimination between two respective pulses incident from the GST side and the GCT side. We used the finite-difference time-domain method to calculate the electric field inside the GST and GCT layers. Figure 3(a) illustrates a simulation model, in which the pulse is incident from the GCT side. A light source modeled as a linearly polarized Gaussian beam with a wavelength of 800 nm, which is the wavelength of the femtosecond pulsed laser for amorphization. The distribution of electric field intensity inside the crystalline-GCT layer is expected to be significantly different depending on the phase of the GST layer. This causes a difference in the amount of absorption of the pulse energy (ΔF defined above) and thus a difference in the resultant phase (degree of amorphization) of the GCT layer. The reason that the GCT layer is thicker than the GST layer is because the contrast in the refractive indices of the crystalline and amorphous phases of GCT is smaller than that of GST; therefore, the difference is partially compensated by making the GCT layer thicker.

 

Fig. 3 Calculation of the electric field intensity distribution in the GST/GCT double layer structure for a pulse incident from the (a) GCT side and (c) GST side. Distribution of electric field intensity along the vertical direction z, for a pulse incident from the (b) GCT side and (d) GST side. Amo: amorphous, Cry: crystalline.

Download Full Size | PDF

Figure 3(b) shows the distribution of electric field intensity in the GST and GCT layers. Black and red curves are obtained by assuming the GST to be crystalline and amorphous, respectively. The interference of incident and reflected light generates a significant difference in the field distribution due to the distinct contrast in the refractive index of GST; a larger field is generated inside the crystalline GCT layer in combination with crystalline GST because of the larger refractive index (reflectivity) of the crystalline phase. The absorption by the GCT layer for the GST layer in the crystalline phase and in the amorphous phase is estimated to be 53% and 39%, respectively, and therefore ΔF = 14%. By providing an optical pulse with appropriate fluence, amorphization of the GCT layer occurs when the GST is in the crystalline phase, while only partial amorphization occurs when GST is in the amorphous phase.

In contrast, the field distribution in the GST layer was calculated with respect to the phase of GCT (Fig. 3(c)) for a pulse incident from the opposite side. In this case, as shown in Fig. 3(d), the field intensity in the GST layer is larger when the GCT layer is in the amorphous phase, which exhibits larger reflectivity than the crystalline phase. Although, as expected, an opposite tendency to that in Fig. 3(b) was obtained, the difference between the cases of crystalline GCT and amorphous GCT in Fig. 3(d) was less significant compared to that in Fig. 3(b) because the contrast in the refractive indices of crystalline and amorphous GCT is much smaller than GST. The results in Figs. 3(b) and 3(d) quantitatively support the principle of the pulse timing discriminator depicted in Fig. 1.

4. Experimental demonstration

As an experimental demonstration, the GST/GCT layered structure shown schematically in Fig. 4(a) was prepared by sputter deposition. A SiO₂ layer was inserted between the GST and GCT layers to avoid intermixing during the annealing process for crystallization as an initial state. The femtosecond pulsed laser was divided into two (counter-propagating) paths, one of which was incident from the GST side and the other from the GCT side of the sample. For both paths, the laser beams were focused onto the GST and GCT layers using a microscope objective lens with a numerical aperture of 0.15. The focal points of the two beams were carefully aligned and completely overlapped. A time delay Δt, was also set between the two pulses, which was defined as Δt>0 when the pulse incident from the GST side preceded that from the GCT side.

 

Fig. 4 (a) Schematic of sample structure. (b) CLSM observation of amorphous marks from the GST side. Amorphous marks were created by two counter-propagating femtosecond pulses with time delays ranging from Δt = −10 ps to Δt = + 10 ps. Δt is defined as a positive delay when the pulse was incident from the GST side first. The fluences of pulses incident from the GST and GCT sides were 34 and 46 mJ/cm², respectively. (c) Same as in (b) except the observation was performed from the GCT side and the fluence of both pulses was 38 mJ/cm². The contrast of the images in (b) and (c) is enhanced to clearly show the features of interest.

Download Full Size | PDF

On the basis of the numerical calculation in the previous chapter, the operation principle of phase-change discriminator can be summarized as follows: (1) For ∆t<0, the first pulse is incident from the GCT side and the GCT layer is amorphized. The fluence of the first pulse should be larger than the threshold for amorphization of the GCT layer but not so large as to amorphize the GST layer. The second pulse incident from the GST side amorphizes the GST layer. The fluence of the second pulse is set in a similar way to the first pulse. (2) For ∆t>0, the first pulse is incident from the GST side and the GST layer is amorphized because the electric field intensity in the GST layer is almost the same for GCT layer in the crystalline phase and in the amorphous phase (see Fig. 3(d)). The second pulse incident from the GCT side does not completely amorphize the GCT layer (The GCT layer remains crystalline or is only partially amorphized) because the electric field intensity in the GCT layer is much lower for the GST layer in the amorphous phase than in the crystalline phase (see Fig. 3(b)).

Figure 4(b) shows the amorphous marks observed from the GST side for pulse delays of −10 ps<Δt< + 10 ps with 2 ps intervals. The pulse fluences incident from the GST and GCT sides were set to be 34 and 46 mJ/cm², respectively. For Δt<0, the amorphous mark appeared as a darker spot and it was diminished at Δt>0. As illustrated in Fig. 1 and theoretically confirmed in Fig. 3, the difference in the mark contrast for Δt<0 and for Δt>0 is due to the difference in the degree of amorphization of GCT. For Δt>0, where the GST layer was in the amorphous phase, the GCT layer was less amorphized by the second pulse compared to that for Δt<0, where the GST layer is in the crystalline phase, the GCT was more amorphized by the first pulse.

Figure 4(c) shows amorphous marks observed from the GCT side for the same delays presented in Fig. 4(b). The fluences of both pulses were set to 38 mJ/cm². For Δt<0, the amorphous mark appeared as brighter spots, which then became darker spots at Δt>0. This result can be attributed to the difference in the degree of amorphization of GCT, which is qualitatively the same mechanism as described for GST. For Δt<0, the brighter contrast is due to the full amorphization of GCT by the first pulse because GST was in the crystalline phase and this causes a stronger electric field intensity in the GCT layer, as shown in Fig. 3(b). For Δt>0, the GST layer was fully amorphized by the first pulse, of which the fluence was larger than that in the case of Fig. 4(b). This significantly reduces the field intensity of the second pulse in the GCT layer below the threshold for amorphization of GCT and the GCT layer remains in the crystalline phase, which has lower reflectivity than that of the amorphous phase. This is the reason why the amorphous mark has darker contrast at Δt>0. A picosecond temporal resolution was thus obtained as a result of sub-picosecond nonthermal amorphization.

5. Discussion

To quantitatively evaluate the observed contrast, the reflectivity from the double layer was calculated for three different combinations of the GST and GCT phases. The configuration was the same as that shown in Fig. 3(a), which corresponds to observation from the GCT side. The wavelength is assumed to be 408 nm, which is the wavelength of the laser diode mounted in the CLSM. The background (the initial state) corresponds to the combination of crystalline-GCT/crystalline-GST. For the amorphous marks at Δt<0 and Δt>0, the combinations are amorphous-GCT/amorphous-GST and crystalline-GCT/amorphous-GST, respectively. The reflectivity of the background, the mark at Δt<0, and the mark at Δt>0, were 31.0, 33.9, and 30.3%, respectively. This result is in agreement with the experimental result.

The phase-change discriminator we proposed is inferior to SOAs or other nonlinear optical devices in terms of functionality, accuracy and stability. Regarding the functionality, for example, SOAs can provide a response which more closely resemble the STDP characteristics in biological neuron. The phase-change discriminator, however, is promising as a nonvolatile energy-efficient and compact component, which is essentially important for future applications to extremely large-scale neural networks mimicking the human brain. Nonthermal amorphization of GST and GCT, in which the principal structural change is completed in a sub-picosecond time scale after femtosecond pulse excitation, is also advantageous to develop a discriminator with a higher temporal resolution. This will be crucial for building ultrafast neuromorphic computing. In principle, the ultrafast response enables to discriminate the pulse timing with a temporal resolution < 1 ps.

6. Summary

To summarize, the functionality of a pulse delay discriminator was implemented with a GST/GCT double layer structure. By exploiting the opposite refractive index change behavior of GST and GCT upon phase change, the arrival order of two counter-propagating pulses was coded into the degree of amorphization of the GCT layer. By carefully tuning the pulse fluence, amorphous marks were experimentally obtained with opposite contrast (with respect to the crystalline background) between the positive and negative delay of the two pulses.