High-stability large-window all-optical phase change brain-like computing material and device

Abstract

The invention discloses a high-stability large-window all-optical phase change brain-like computing material and device. The device structurally comprises a substrate, an optical waveguide layer, a dielectric layer, a phase change layer and a top anti-oxidation layer in sequence from bottom to top, the device phase change layer is made of germanium-enriched germanium-antimony-tellurium phase change materials, the chemical formula is GexSbyTez, x is larger than or equal to 25 and smaller than or equal to 50, y is larger than or equal to 15 and smaller than or equal to 25, and z is larger than or equal to 30 and smaller than or equal to 50. The phase change material has the advantages of high crystallization temperature, good amorphous thermal stability, large optical property difference between amorphous phases and crystalline phases, large switching ratio of the device, and excellent cycle stability.

Description

Technical Field

[0001] This invention relates to the fields of phase change materials and neuromorphic in-memory computing technology, specifically to a highly stable, large-window, all-optical phase change brain-like computing material and device. Background Technology

[0002] All-optical storage and neuromorphic computing chips based on phase change materials (PCMs) represent one of the most promising frontiers in the information field. They combine the non-volatility of PCMs with the ultra-high speed, high bandwidth, and low crosstalk advantages of photonics, providing a novel physical platform for overcoming the "Von Neumann bottleneck" and "memory wall" of traditional electronic computing. PCMs exhibit rapid and reversible phase transitions between amorphous and crystalline phases and possess non-volatility, meaning their state remains stable after a phase transition triggered by external stimuli (light pulses), requiring no continuous power supply and supporting ultra-low static power consumption. Furthermore, PCMs exhibit significant differences in optical properties between their amorphous and crystalline phases, possessing extremely strong optical modulation capabilities and achieving high optical contrast even at the nanoscale. Their excellent stability and optical modulation capabilities provide a physical foundation for high-density data storage and high-precision analog computing.

[0003] Currently, high-temperature processes in embedded chip manufacturing, the service temperature range (-40~165 ℃) of edge computing, and practical high-precision multi-level in-memory computing applications place stringent demands on phase change materials, including high thermal stability, large extinction ratio, long lifetime, and cycle stability. However, the germanium-antimony-tellurium phase change material currently used in commercial chips has a low crystallization temperature (~150 ℃) and poor amorphous thermal stability, making it impossible to avoid the possibility of thermally induced crystallization during preparation or use. Furthermore, the extinction ratio of germanium-antimony-tellurium reported in existing studies is only about 0.5 dB, and it can only be maintained for 100 cycles, exhibiting poor cycle stability, making it difficult to achieve precise and stable modulation of multiple intermediate states in practical applications.

[0004] Therefore, there is an urgent need to develop a novel all-optical phase change material with high thermal stability, large optical window and long cycle life to meet the practical application requirements of embedded chip processing and edge computing. Summary of the Invention

[0005] To overcome the shortcomings of the existing technology, this invention provides a highly stable, large-window all-optical phase-change neuromorphic computing material and device. Based on germanium-enriched germanium-antimony-tellurium phase-change material, the enriched germanium element increases the proportion of strong covalent bonds in the amorphous structure to a certain extent, improving the amorphous thermal stability of the material. At the same time, it increases the chemical bonding and structural differences between the amorphous and crystalline phases, and increases the differences in optical properties between the amorphous and crystalline phases. Based on this, the all-optical neuromorphic computing device has the advantages of high thermal stability, large on / off ratio, and good cycle stability.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A highly stable, large-window, all-optical phase-change neuromorphic computing material, wherein the material is a germanium-enriched germanium-antimony-tellurium phase-change material with the chemical formula Ge. x Sb y Te z Where 25≤x≤50, 15≤y≤25, 30≤z≤50, x+y+z=100, and x, y, and z are the atomic percentages of the elements; The thickness of the phase change layer composed of the germanium-enriched germanium-antimony-tellurium phase change material ranges from 5 to 50 nm. The germanium-enriched germanium-antimony-tellurium phase change material has an amorphous phase bonded by covalent bonds and a crystalline phase bonded by metallic covalent bonds, with both phases having an octahedral structure as the main feature.

[0007] Germanium doping increases the proportion of tetrahedral configurations in the amorphous phase, making the chemical bonding and structural differences between the two phases more significant, resulting in more obvious differences in their optical properties and greater optical contrast.

[0008] The germanium-enriched germanium-antimony-tellurium phase change material exhibits the strongest modulation capability in the 1530-1565 nm C-band of optical fiber communication, which is commonly used in optical fiber communication; at the center wavelength of 1550 nm, the refractive index difference Δ between the amorphous and crystalline phases is the largest. n ≥1.7, extinction coefficient difference Δ k ≥0.8.

[0009] An all-optical phase-change neuromorphic computing device based on germanium-enriched germanium-antimony-tellurium phase change material is presented. The amorphous phase has a low optical constant value, low absorption loss, and high transmittance. The crystalline phase has a high optical constant value, high absorption loss, and low transmittance. The device has a large transmittance window at a wavelength of 1550 nm and an extinction ratio exceeding 3 dB. It can be stably cycled for more than 10,000 times at this extinction ratio with a cycle window fluctuation of ≤8%, demonstrating excellent cycle stability.

[0010] The all-optical phase change neuromorphic computing device based on germanium-enriched germanium-antimony-tellurium phase change material includes, from bottom to top, a substrate, an optical waveguide layer, a phase change layer, and a top anti-oxidation layer along the Z-axis. The phase change layer is a germanium-enriched germanium-antimony-tellurium phase change material; The optical waveguide layer includes a lower square structure and a ridge-shaped central protrusion on the surface of the square structure. The ridge-shaped central protrusion is arranged along the center of the X-axis of the square structure surface and has the same length as the square structure. A phase change layer is disposed on the upper surface of the ridge-shaped central protrusion, and the width of the phase change layer is the same as the width of the ridge-shaped central protrusion. The width of the optical waveguide layer and the substrate is 2~10 μm; The thickness of the phase change layer ranges from 5 to 50 nm; The thickness of the top anti-oxidation layer ranges from 5 to 50 nm; The length w of the phase change layer and the top anti-oxidation layer ranges from 40 to 80 μm; The length of the optical waveguide layer of the device ranges from 200 to 500 μm, the width of the square structure ranges from 300 to 800 nm, and the thickness ranges from 150 to 500 nm. The phase change layer is a germanium-enriched germanium-antimony-tellurium phase change material with a crystallization temperature exceeding 250 ℃ and can maintain stable data for more than ten years at a high temperature of 209 ℃, exhibiting excellent amorphous thermal stability.

[0011] The device substrate material is silicon dioxide, the optical waveguide layer material is either silicon or silicon nitride, and the top anti-oxidation layer is indium tin oxide (ITO).

[0012] The preparation process of the germanium-enriched germanium-antimony-tellurium phase change material is one of magnetron sputtering, electron beam evaporation, chemical vapor deposition, laser pulse deposition, and atomic layer deposition.

[0013] A 1550 nm pump laser was used to apply single-step write pulses and double-step erase pulses to the device, switching the phase state of the phase change material. The applied photocurrent ranged from 50 to 600 mA, the write pulse width ranged from 50 to 300 ns, and the erase pulse width ranged from 800 to 2000 ns. After applying the pulses, a 1540 nm probe laser was used simultaneously to detect the device signal value.

[0014] The neuromorphic devices described can be applied to several core application areas, such as high-precision simulated synapse systems, physical storage pools in storage computing, multi-valued photonic memories in high-bandwidth caching, and optoelectronic hybrid in-memory computing units.

[0015] The beneficial effects of this invention are: This invention, by increasing the germanium content instead of using other elements for doping, achieves stronger chemical bond compatibility with the parent material germanium-antimony-tellurium. Furthermore, the germanium element increases the strong covalent bonds of germanium-tellurium and the germanium-germanium homovalent bonds in the amorphous phase, improving the overall rigidity and average bond strength of the amorphous network and enhancing the amorphous thermal stability of the material. Through elemental regulation, the material's crystallization temperature exceeds 250℃, more than 60% higher than that of typical commercial phase change materials like germanium-antimony-tellurium (~150℃). This invention provides a solution to the problem of device failure under high-temperature processes in chip manufacturing back-end processes and the performance stability of edge devices under extreme operating temperatures, broadening the temperature application scenarios for such phase change neuromorphic computing chips and devices.

[0016] In this invention, the germanium-antimony-tellurium amorphous phase is covalently bonded, while the crystalline phase is bonded by metallic covalent bonds. Both phases are predominantly octahedral in structure. Germanium doping increases the proportion of tetrahedral structures in the amorphous phase, making the chemical bonding and structural differences between the two phases more significant, resulting in more pronounced differences in their optical properties and greater optical contrast. The amorphous phase structure is disordered with a wide band gap, exhibiting a weak polarization response to light, manifested as a low refractive index and extinction coefficient. The crystalline phase structure is periodically arranged with a very small band gap, facilitating carrier transitions and resulting in a strong polarization response and absorption of light, manifested as a high refractive index and extinction coefficient. By controlling the germanium content, the optical constant Δ at 1550 nm of the germanium-antimony-tellurium material can be optimized. n ≥1.7, extinction coefficient difference Δ k ≥0.8, Δ before doping n ~1.3, Δ k ~0.5.

[0017] The device of this invention exhibits an extinction ratio exceeding 3 dB and can stably cycle for over 10,000 times within this order of magnitude window, with a cycle window fluctuation of ≤8%, demonstrating superior on / off ratio and cycle stability. Furthermore, the memory device provided by this invention has a simple and compact structure, achieving a high extinction ratio at the nanoscale volume. Its fabrication process is compatible with existing CMOS processes, facilitating large-scale integrated production. Attached Figure Description

[0018] Figure 1 The ten-year data retention curve of the germanium-enriched germanium-antimony-tellurium phase change material in Example 3.

[0019] Figure 2 The refractive index difference Δ between the crystalline and amorphous phases of the germanium-enriched germanium-antimony-tellurium phase change material in Example 3 is shown. n Curve showing the change with wavelength.

[0020] Figure 3 The difference Δ between the crystalline and amorphous extinction coefficients of the germanium-enriched germanium-antimony-tellurium phase change material in Example 3 is shown in Figure 3. k Curve showing the change with wavelength.

[0021] Figure 4 This is a schematic diagram of the structure of an all-optical neuromorphic computing device based on germanium-enriched germanium-antimony-tellurium phase change material. Detailed Implementation

[0022] The present invention will now be described in further detail with reference to the accompanying drawings.

[0023] like Figure 4 As shown, this invention discloses a highly stable, large-window, all-optical phase-change neuromorphic computing material and device. The device has four layers from bottom to top: 1-substrate, 2-optical waveguide layer, 3-phase change layer and 4-top anti-oxidation layer. The optical waveguide layer 2 includes a lower square structure and a ridge-shaped central protrusion on the surface of the square structure. The ridge-shaped central protrusion is arranged along the center of the X-axis of the square structure surface and has the same length as the square structure. A phase change layer 3 is provided on the upper surface of the ridge-shaped central protrusion, and the width of the phase change layer 3 is the same as the width of the ridge-shaped central protrusion. The widths of the optical waveguide layer 2 and the substrate 1 are 2~10 μm; The thickness d of phase change layer 3 ranges from 5 to 50 nm; The thickness of the top anti-oxidation layer 4 ranges from 5 to 50 nm; The length w of the phase change layer 3 and the top anti-oxidation layer 4 ranges from 40 to 80 μm.

[0024] The optical waveguide layer has a length ranging from 200 to 500 μm, a square structure width ranging from 300 to 800 nm, and a thickness ranging from 150 to 500 nm.

[0025] A phase change layer 3 of phase change material is deposited in the middle region of the optical waveguide layer 2 using photolithography, and a top anti-oxidation layer 5 is deposited on top of the phase change layer 3 to prevent oxidation of the phase change layer.

[0026] The present invention will be further illustrated below with specific embodiments.

[0027] Example 1 This embodiment is based on Ge 25 Sb 25 Te 50 The process of an all-optical neuromorphic computing device using phase change materials is as follows: (1) Thin film preparation: A germanium-enriched germanium-antimony-tellurium thin film was deposited on a silicon wafer by co-sputtering with two targets, Ge and GeSb2Te4, using argon gas to start the ignition. Sputtering parameters: sputtering pressure was 3 mTorr, argon gas flow rate was 15 sccm; DC power supply was used for all targets, and the atomic percentage of the thin film and sputtering rate were controlled by adjusting the power of the Ge target and GeSb2Te4 target (10~60 W).

[0028] (2) Thin film performance testing: The resistance of the thin film was measured using a dual-channel source meter and a two-point probe method. The thin film was annealed at 400 °C using an in-situ heating and cooling stage. The refractive index and extinction coefficient of the amorphous and crystalline phases of the thin film were tested using an elliptic polarization spectrometer.

[0029] (3) Device design: Substrate 1 is SiO2; optical waveguide layer 2 is Si; phase change layer 3 is Ge 25 Sb 25 Te 50 The top anti-oxidation layer 4 is made of ITO transparent material.

[0030] (4) Device testing: A 1550nm pump laser was used to apply a single-step write pulse and a double-step erase pulse to the device, switching the phase state of the phase change material. The applied photocurrent range was 50mA, the write pulse pulse width range was 50ns, and the erase pulse pulse width range was 800ns. After applying the pulse, a 1540nm probe laser was used simultaneously to detect the device signal value.

[0031] Example 2 This embodiment is based on Ge 43 Sb 19 Te 38 The process of an all-optical neuromorphic computing device using phase change materials is as follows: (1) Thin film preparation: A germanium-enriched germanium-antimony-tellurium thin film was deposited on a silicon wafer by magnetron sputtering-physical vapor deposition with argon gas ignition and co-sputtering using a Ge target and a GeSb2Te4 target. Sputtering parameters: sputtering pressure was 3 mTorr, argon gas flow rate was 15 sccm; DC power supply was used for all power supplies, and the atomic percentage of the thin film and sputtering rate were controlled by adjusting the power of the Ge target and GeSb2Te4 target (10~60 W).

[0032] (2) Thin film performance testing: The resistance of the thin film was measured using a dual-channel source meter and a two-point probe method. The thin film was annealed at 400 °C using an in-situ heating and cooling stage. The refractive index and extinction coefficient of the amorphous and crystalline phases of the thin film were tested using an elliptic polarization spectrometer.

[0033] (3) Device design: Substrate 1 is SiO2; optical waveguide 2 is Si; phase change layer 3 is Ge 43 Sb 19 Te 38 The top anti-oxidation layer 4 is made of ITO transparent material.

[0034] (4) Device Testing: A 1550nm pump laser was used to apply single-step write pulses and double-step erase pulses to the device, switching the phase state of the phase change material. The applied photocurrent ranged from 600 mA, the write pulse pulse width ranged from 300 ns, and the erase pulse pulse width ranged from 2000 ns. After applying the pulses, a 1540 nm probe laser was used simultaneously to detect the device signal value. Specific Application Description: This highly stable, large-window all-optical phase change neuromorphic device can be applied to high-precision synapse simulation systems. By utilizing different refractive index states of the phase change material (amorphous low refractive index / crystalline high refractive index), the weights of biological synapses are simulated to construct an artificial neural network. Its large operating window and good cycle life ensure that the weights can be precisely and stably controlled (trained) and read (inferred) tens of thousands of times, enabling the storage and calculation of weights in deep neural networks and convolutional neural networks.

[0035] Example 3 This embodiment is based on Ge 50 Sb 15 Te 35 The process of an all-optical neuromorphic computing device using phase change materials is as follows: (1) Thin film preparation: A germanium-enriched germanium-antimony-tellurium thin film was deposited on a silicon wafer by magnetron sputtering-physical vapor deposition with argon gas ignition and co-sputtering using a Ge target and a GeSb2Te4 target. Sputtering parameters: sputtering pressure was 3 mTorr, argon gas flow rate was 15 sccm; DC power supply was used for all power supplies, and the atomic percentage of the thin film and sputtering rate were controlled by adjusting the power of the Ge target and GeSb2Te4 target (10~60 W).

[0036] (2) Thin film performance testing: The resistance of the thin film was measured using a dual-channel source meter and a two-point probe method. The thin film was annealed at 400 °C using an in-situ heating and cooling stage. The refractive index and extinction coefficient of the amorphous and crystalline phases of the thin film were tested using an elliptic polarization spectrometer.

[0037] (3) Device design: Substrate 1 is SiO2; optical waveguide 2 is Si; phase change layer 3 is Ge 50 Sb 15 Te 35 The top anti-oxidation layer 4 is made of ITO transparent material.

[0038] (4) Device Testing: A 1550nm pump laser was used to apply single-step write pulses and double-step erase pulses to the device, switching the phase state of the phase change material. The applied photocurrent ranged from 600 mA, the write pulse pulse width ranged from 300 ns, and the erase pulse pulse width ranged from 2000 ns. After applying the pulses, a 1540 nm probe laser was used simultaneously to detect the device signal value. Specific Application Description: This highly stable, large-window all-optical phase change neuromorphic device can be applied to high-precision synapse simulation systems. By utilizing different refractive index states of the phase change material (amorphous low refractive index / crystalline high refractive index), the weights of biological synapses are simulated to construct an artificial neural network. Its large operating window and good cycle life ensure that the weights can be precisely and stably controlled (trained) and read (inferred) tens of thousands of times, enabling the storage and calculation of weights in deep neural networks and convolutional neural networks.

[0039] like Figure 1 As shown, this is a germanium-enriched Ge. 50 Sb 15 Te 35Ten-year data retention curves were obtained. Three temperatures—270 °C, 275 °C, and 280 °C—were selected before reaching the crystallization temperature. The temperature was increased at a rate of 10 °C / min to the corresponding temperature and held at that temperature. The film resistance was recorded in real time, and the time it took for the resistance to drop to half of its initial value was defined as the failure time at that temperature. t By analyzing ln t —1 / K B Linear fitting of T yielded a failure time of ten years (3.15 × 10⁻⁶). 8 The temperature corresponding to s) is 170 ℃.

[0040] like Figure 2 As shown, this is a germanium-enriched Ge. 50 Sb 15 Te 35 The refractive index difference Δ between crystalline and amorphous phases n The refractive index of the amorphous state varies with wavelength. Within the wavelength range of 200–650 nm. n Greater than that of the crystalline state; in the wavelength range of 650 ~ 2100 nm, the refractive index of the amorphous state is... n Smaller than crystalline. This can be determined by the difference in refractive index Δ between the amorphous and crystalline phases. n To predict the window size in the device. At the center wavelength λ = 1550 nm in the communication band, Δ n ≥1.7.

[0041] like Figure 3 As shown, this is a germanium-enriched Ge. 50 Sb 15 Te 35 The difference in extinction coefficients between amorphous and crystalline phases Δ k The extinction coefficient of the amorphous state varies with wavelength. Within the wavelength range of 200–260 nm. k Greater than that of the crystalline state; in the wavelength range of 260 ~ 2100 nm, the extinction coefficient of the amorphous state is... k Smaller than crystalline. This can be determined by the difference Δ between the extinction coefficients of the amorphous and crystalline phases. k To predict the window size in the device. At the center wavelength λ=1550 nm in the communication band, the Δ between the amorphous and crystalline phases... k ≥0.8.

Claims

1. A highly stable, large-window, all-optical phase-change neuromorphic computing material, characterized in that, The material is a germanium-enriched germanium-antimony-tellurium phase change material with the chemical formula Ge. x Sb y Te z Where 25≤x≤50, 15≤y≤25, 30≤z≤50, x+y+z=100, and x, y, and z are the atomic percentages of the elements; The germanium-enriched germanium-antimony-tellurium phase change material has an amorphous phase bonded by covalent bonds and a crystalline phase bonded by metallic covalent bonds, with both phases having an octahedral structure as the main feature.

2. The highly stable, large-window, all-optical phase-change neuromorphic computing material according to claim 1, characterized in that, The germanium-enriched germanium-antimony-tellurium phase change material exhibits the strongest modulation capability in the 1530-1565 nm C-band of optical fiber communication, which is commonly used in optical fiber communication; at the center wavelength of 1550 nm, the refractive index difference Δ between the amorphous and crystalline phases is the largest. n ≥1.7, extinction coefficient difference Δ k ≥0.

8.

3. A fully optical phase-change neuromorphic computing device based on germanium-enriched germanium-antimony-tellurium phase-change material, characterized in that, Along the Z-axis, from bottom to top, it includes a substrate, an optical waveguide layer, a phase change layer, and a top anti-oxidation layer; The phase change layer is the germanium-enriched germanium-antimony-tellurium phase change material as described in claim 1 or 2; The optical waveguide layer includes a lower square structure and a ridge-shaped central protrusion on the surface of the square structure. The ridge-shaped central protrusion is arranged along the center of the X-axis of the square structure surface and has the same length as the square structure. A phase change layer is disposed on the upper surface of the ridge-shaped central protrusion, and the width of the phase change layer is the same as the width of the ridge-shaped central protrusion. The width of the optical waveguide layer and the substrate is 2~10 μm; The thickness of the phase change layer ranges from 5 to 50 nm; The thickness of the top anti-oxidation layer ranges from 5 to 50 nm; The length w of the phase change layer and the top anti-oxidation layer ranges from 40 to 80 μm; the length of the optical waveguide layer ranges from 200 to 500 μm, the width of the square structure ranges from 300 to 800 nm, and the thickness ranges from 150 to 500 nm.

4. The all-optical phase-change neuromorphic computing device based on germanium-enriched germanium-antimony-tellurium phase-change material according to claim 3, characterized in that, The substrate material is silicon dioxide, the optical waveguide layer material is either silicon or silicon nitride, and the top anti-oxidation layer is indium tin oxide (ITO).

5. The all-optical phase-change neuromorphic computing device based on germanium-enriched germanium-antimony-tellurium phase-change material according to claim 3, characterized in that, The preparation process of the germanium-enriched germanium-antimony-tellurium phase change material is one of magnetron sputtering, electron beam evaporation, chemical vapor deposition, laser pulse deposition, and atomic layer deposition.

6. The all-optical phase-change neuromorphic computing device based on germanium-enriched germanium-antimony-tellurium phase-change material according to claim 3, characterized in that, A 1550 nm pump laser was used to apply single-step write pulses and double-step erase pulses to the device, switching the phase state of the phase change material. The applied photocurrent ranged from 50 to 600 mA, the write pulse width ranged from 50 to 300 ns, and the erase pulse width ranged from 800 to 2000 ns. After applying the pulses, a 1540 nm probe laser was used simultaneously to detect the device signal value.

7. The application of an all-optical phase-change neuromorphic computing device based on germanium-enriched germanium-antimony-tellurium phase-change material according to any one of claims 3-6, characterized in that, The neuromorphic devices are specifically applied to high-precision simulated synapse systems, physical storage pools in storage computing, multi-valued photonic memories in high-bandwidth caches, and optoelectronic hybrid in-memory computing units.

Images