Inp optoelectronic chip integrating single-carrier photodetector-based neurons

Abstract

The present disclosure relates to the technical field of optical communication, and particularly relates to an InP optoelectronic chip integrating a neuron based on a single-carrier photodetector. The neuron comprises: a nonlinear activation function unit, the nonlinear activation function unit comprising at least one pair of single-carrier photodetectors, any pair of single-carrier photodetectors comprising a first single-carrier photodetector and a second single-carrier photodetector; wherein a first saturation photocurrent corresponding to the first single-carrier photodetector and a second saturation photocurrent corresponding to the second single-carrier photodetector are different, and the nonlinear activation function unit is configured to realize the function of an activation function according to a corresponding relationship between a difference between photocurrents and light intensity, wherein the difference between photocurrents is a difference between a first photocurrent output by the first single-carrier photodetector and a second photocurrent output by the second single-carrier photodetector. The present disclosure can reduce energy consumption and improve computing speed.

Description

Technical Field

[0001] This disclosure relates to the field of optical communication technology, and more particularly to an InP optoelectronic chip integrating neurons based on a single-row carrier photodetector. Background Technology

[0002] In optoelectronic hybrid neural networks, photonic devices are primarily used for weight allocation, while electronic devices are used to implement nonlinear activation functions. The performance of a neural network largely depends on the performance of its nonlinear activation function. Therefore, concentrating all linear and nonlinear operations in the optical domain can effectively improve the efficiency of optical computing.

[0003] In related technologies, schemes for implementing nonlinear activation functions using photonic devices can be divided into all-optical schemes and photoelectric conversion schemes. All-optical schemes require sufficiently strong nonlinear cross-phase modulation effects to achieve nonlinearity, thus increasing energy consumption due to the need for high input optical power. Photoelectric conversion schemes, on the other hand, require the introduction of additional delay lines to allow the optical and electrical signals to enter the photonic device simultaneously, which limits the computational speed and increases system complexity. Summary of the Invention

[0004] This disclosure provides an InP optoelectronic chip that integrates neurons based on a single-row carrier photodetector, with the main purpose of reducing energy consumption and increasing computing speed.

[0005] According to one aspect of this disclosure, an InP optoelectronic chip integrating a neuron based on a single-row carrier photodetector is provided. The neuron includes: a nonlinear activation function unit, which includes at least one pair of single-row carrier photodetectors, wherein any one pair of single-row carrier photodetectors includes a first single-row carrier photodetector and a second single-row carrier photodetector; wherein...

[0006] The first saturated photocurrent corresponding to the first single-row carrier photodetector and the second saturated photocurrent corresponding to the second single-row carrier photodetector are different. The nonlinear activation function unit is used to realize the function of activation function according to the correspondence between the difference of photocurrent and the light intensity. The difference of photocurrent is the difference between the first photocurrent output by the first single-row carrier photodetector and the second photocurrent output by the second single-row carrier photodetector.

[0007] Optionally, the first saturated light intensity corresponding to the first single-row carrier photodetector is greater than the second saturated light intensity corresponding to the second single-row carrier photodetector. The correspondence between the difference in photocurrent and the light intensity includes: When the light intensity is less than the second saturation light intensity, the first responsivity corresponding to the first single-row carrier photodetector and the second responsivity corresponding to the second single-row carrier photodetector are the same, and the difference in photocurrent is zero. When the light intensity is not less than the second saturation light intensity and less than the first saturation light intensity, the first responsivity remains unchanged, the second responsivity is zero, the second photocurrent maintains the second saturation photocurrent unchanged, and the difference in photocurrent is proportional to the light intensity. When the light intensity is not less than the first saturation light intensity, both the first responsivity and the second responsivity are zero, the first photocurrent remains unchanged at the first saturation photocurrent, the second photocurrent remains unchanged at the second saturation photocurrent, and the difference between the photocurrents is a constant value.

[0008] Optionally, the absorption layer thickness of the first single-row carrier photodetector and the second single-row carrier photodetector is the same, so that when the light intensity is less than the second saturation light intensity, the first responsivity of the first single-row carrier photodetector and the second responsivity of the second single-row carrier photodetector are the same. The first single-row carrier photodetector and the second single-row carrier photodetector have different structural dimensions or other epitaxial layer doping concentrations, except for the absorption layer thickness, so that the first saturated photocurrent and the second saturated photocurrent are different.

[0009] Optional, other structural dimensions include the thickness and doping concentration of the cliff layer, the thickness and doping concentration of the collection layer, the ratio of the thickness of the depletion absorber layer to the thickness of the non-depletion absorber layer, and the doping concentration.

[0010] Optionally, the neuron based on a single-row carrier photodetector also includes a light source, a linear computation unit, and an output unit; wherein, The light source is connected to the input terminal of the linear computing unit and is used to output at least two first light signals to the linear computing unit; The output of the linear calculation unit is connected to the input of the nonlinear activation function unit, and is used to perform linear calculations on at least two first light signals input from the light source to obtain and output a second light signal to the nonlinear activation function unit. The output of the nonlinear activation function unit is connected to the output unit and is used to divide the second optical signal into at least two third optical signals, and convert the third optical signals into current signals according to the single-row carrier photodetector. The output unit is used to convert the current difference signal between at least one pair of single-row carrier photodetectors into a voltage signal and output the voltage signal.

[0011] Optionally, the light source includes at least two semiconductor lasers, and the semiconductor lasers include at least one of the following: Distributed feedback semiconductor laser; Distributed Bragg semiconductor laser; Vertical cavity surface-emitting semiconductor laser.

[0012] Optionally, the linear computing unit includes at least two first amplitude modulators, at least two second amplitude modulators, a wavelength division multiplexer, and a semiconductor laser, with each of the first and second amplitude modulators corresponding to one another; wherein, The input terminal of the first amplitude modulator is connected to the output terminal of the semiconductor laser, and is used to perform amplitude modulation on the first optical signal according to the input signal to obtain the first sub-optical signal; The input of the second amplitude modulator is connected to the output of the first amplitude modulator, and is used to perform amplitude modulation on the first sub-optical signal according to the weight signal to obtain the second sub-optical signal. The input of the wavelength division multiplexer is connected to the output of at least two second amplitude modulators to perform multiplexing on at least two second sub-optical signals to obtain a second optical signal.

[0013] Optionally, both the first amplitude modulator and the second amplitude modulator include at least one of the following: Electroabsorption modulator; Mach-Zehnder modulator; Micro-ring modulator.

[0014] Optionally, the nonlinear activation function unit includes a multimode interferometer; wherein, The input of the multimode interferometer is connected to the output of the linear computing unit to perform equal-splitting processing on the second optical signal to obtain at least two third optical signals; The input terminal of the single-carrier photodetector is connected to the output terminal of the multimode interferometer to convert the third optical signal into a current signal; The output terminal of the single-row carrier photodetector is connected to the output unit.

[0015] Optionally, the output unit includes a transimpedance amplifier, wherein, The input of the transimpedance amplifier is connected to the output of the nonlinear activation function unit.

[0016] In summary, the InP optoelectronic chip integrating a neuron based on a single-row carrier photodetector provided in this embodiment includes a nonlinear activation function unit. The nonlinear activation function unit includes at least one pair of single-row carrier photodetectors, and any one pair of single-row carrier photodetectors includes a first single-row carrier photodetector and a second single-row carrier photodetector. The first saturated photocurrent corresponding to the first single-row carrier photodetector and the second saturated photocurrent corresponding to the second single-row carrier photodetector are different. The nonlinear activation function unit is used to implement the activation function based on the correspondence between the difference in photocurrent and the light intensity, where the difference in photocurrent is the difference between the first photocurrent output by the first single-row carrier photodetector and the second photocurrent output by the second single-row carrier photodetector. Therefore, by combining the saturated absorption effect of the single-row carrier photodetector in the nonlinear activation function unit, a higher saturated output power and a larger bandwidth can be achieved, enabling the processing of higher-speed signals and reducing the impact of the photoelectric conversion rate on the optical computing speed, thereby improving the computing speed of the neuron. Meanwhile, single-carrier photodetectors can operate under zero bias conditions, resulting in lower energy consumption compared to other solutions, which can reduce the energy consumption of neurons.

[0017] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0018] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein: Figure 1 This is a schematic diagram of the structure of a neuron based on a single-row carrier photodetector provided in an embodiment of this disclosure; Figure 2 This is a schematic diagram illustrating the change of photocurrent with light intensity provided in an embodiment of the present disclosure; Figure 3 This is a schematic diagram illustrating the variation of photocurrent difference with light intensity, provided in an embodiment of this disclosure. Figure 4 This is a schematic diagram of another neuron structure based on a single-row carrier photodetector provided in an embodiment of this disclosure. Detailed Implementation

[0019] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0020] With the development of science and technology, artificial neural networks have achieved tremendous success, finding applications in areas such as facial recognition, autonomous driving, and medical testing. The main driving forces behind artificial neural networks come from three aspects: big data, hardware computing, and algorithms. The rapid development of the data-driven era has resulted in an exponential increase in information and data, providing a vast amount of data samples for artificial neural networks. Furthermore, an increasing number of algorithmic models, such as convolutional neural networks, recurrent neural networks, graph neural networks, and reinforcement learning, have broadened the functionality of neural networks.

[0021] Current research on implementing artificial neural networks using hardware primarily relies on electronic devices, which place high demands on computational speed and accuracy. Traditional methods for implementing neural networks using electronic devices utilize Complementary Metal-Oxide Semiconductor (CMOS) integrated circuits to construct their neurons. However, as the feature size of integrated circuits has approached physical limits in recent years, its development no longer fully conforms to Moore's Law. Furthermore, computer architectures based on the von Neumann architecture face constraints and challenges in terms of both energy consumption and speed when executing models like artificial neural networks that require extensive parallel computation. Therefore, implementing artificial neural networks using electronic devices also presents significant difficulties.

[0022] Photonic devices offer higher transmission rates and greater bandwidth compared to electronic devices, leading to their widespread application in optical communication systems. Furthermore, the superior performance of photons has also been applied to automated optical sensing, spectral imaging, and the reverse engineering of photonic devices. Considering the requirements of neural networks for data parallelization and adaptive processing, photonic devices perfectly match these performance requirements. Therefore, researchers are increasingly focusing on building neural network models based on photonic devices.

[0023] In the early development of photonic neural networks, the difficulty of implementing nonlinear operations on photonic devices led to the proposal of a hybrid optoelectronic approach to construct neural networks. While photonic devices possess many nonlinear properties, the need for each neuron in a neural network to connect with multiple neurons places high demands on their load-bearing capacity. Late 20th-century research on photonic devices could not meet this requirement. Therefore, in hybrid optoelectronic neural networks, photonic devices are primarily used for weight allocation, while electronic devices are used to implement nonlinear activation functions. The performance of a neural network largely depends on the performance of its nonlinear activation function. Therefore, concentrating all linear and nonlinear operations in the optical domain can effectively improve the efficiency of optical computing.

[0024] In related technologies, schemes for realizing nonlinear activation functions using photonic devices can be divided into all-optical schemes and photoelectric conversion schemes. Among them, The all-optical scheme utilizes the gain saturation characteristic of the optical amplifier and induces a cross-phase modulation effect by injecting light into the optical amplifier to achieve nonlinearity. However, since the all-optical scheme requires a sufficiently strong nonlinear cross-phase modulation effect to achieve nonlinearity, the input optical power needs to be high enough, which increases energy consumption. The photoelectric conversion scheme splits the optical signal into two paths: one path serves as the input optical signal for the photonic device, and the other path is converted into an electrical signal by a photodetector, which is then used for the electrical control of the photonic device. Since the propagation speed of optical signals is much faster than that of electrical signals, a delay needs to be introduced into the propagation path of the optical signal so that the optical signal and the electrical signal reach the photonic device simultaneously. Therefore, the computational speed of this scheme is limited, and the complexity of the system is also increased.

[0025] The most suitable materials for active optoelectronic devices are indium phosphide (InP)-based group III-V compounds. Compared to passive devices and silicon (Si)-based materials, group III-V materials have a direct bandgap, resulting in significantly higher light emission and absorption efficiencies than Si. Si-based optoelectronic devices typically require the introduction of other group III-V materials, whose lattice matching with Si is relatively low, limiting fabrication processes. However, in InP-based optoelectronic devices, by adjusting the Ga and As composition in the indium gallium arsenide phosphide (InGaAsP) quaternary compound, various optoelectronic functions can be achieved while maintaining lattice matching with InP. In terms of integration, unlike Si-based optoelectronic integration, InP-based optoelectronic integration can integrate active and passive devices into a microchip using semiconductor processes, truly achieving monolithic integration. Furthermore, compared to traditional devices, InP optoelectronic chips are smaller, lighter, and can be mass-produced.

[0026] In current InP-based photoelectric conversion, there is a demand for photodetectors with increased bandwidth and high output power. However, high output power requires large photocurrent generation, while large bandwidth requires smaller device area, leading to problems such as photogenerated carrier accumulation and limiting power output. Therefore, there is a contradiction between high saturation power and increased bandwidth, which is difficult to achieve simultaneously. Single-row carrier photodetectors (UTC-PDs) can better solve this contradiction.

[0027] The present disclosure will now be described in detail with reference to specific embodiments.

[0028] In the first embodiment, this disclosure provides an InP optoelectronic chip integrating neurons based on a single-row carrier photodetector, such as... Figure 1 As shown, Figure 1 This is a schematic diagram of the structure of a neuron based on a single-row carrier photodetector provided in an embodiment of this disclosure.

[0029] Specifically, the neuron based on a single-row carrier photodetector includes: a nonlinear activation function unit, which includes at least one pair of single-row carrier photodetectors, wherein any pair of single-row carrier photodetectors includes a first single-row carrier photodetector PD1 and a second single-row carrier photodetector PD2; wherein, The first saturated photocurrent corresponding to the first single-row carrier photodetector PD1 is different from the second saturated photocurrent corresponding to the second single-row carrier photodetector PD2. The nonlinear activation function unit is used to realize the function of activation (sigmoid) based on the correspondence between the difference in photocurrent and the light intensity. The difference in photocurrent is the difference between the first photocurrent I1 output by the first single-row carrier photodetector PD1 and the second photocurrent I2 output by the second single-row carrier photodetector PD2.

[0030] According to some embodiments, compared to traditional PIN photodiodes (PDs), uni-traveling-carrier photodiodes (UTC-PDs) can effectively resolve this contradiction, alleviate the space charge effect, achieve higher linearity, and possess both higher saturation current and greater bandwidth. By placing the absorption layer in the P-contact layer near the anode end, the hole transit time can be neglected, and only the electron transit time needs to be considered, thus achieving a frequency response speed of over 100 GHz.

[0031] In some embodiments, a single-row carrier photodetector is a detector employing a UTC-PD. The DC responsivity of a single-row carrier photodetector depends on the thickness of the absorption layer; the smaller the thickness of the absorption layer, the lower the light absorption by the UTC-PD, and the lower the responsivity.

[0032] In some embodiments, under conditions of low incident light power, the photocurrent generated by the single-row carrier photodetector is linearly related to the incident light power. In this case, the single-row carrier photodetector operates in the linear region, and its DC responsivity is equal to the slope of the photoresponse curve. When the incident light power increases, photogenerated electrons and holes cannot reach the contact layers at both ends in time. Photogenerated carriers accumulate in the absorption layer, generating a reverse electric field that hinders electrons from crossing from the absorption layer to the collection layer. Consequently, the response speed of the single-row carrier photodetector slows down, and the photocurrent exhibits a saturation effect.

[0033] It is easy to understand that by incorporating the saturation absorption effect of UTC-PD into the nonlinear activation function unit, a higher optical saturation output power and a larger bandwidth are achieved, enabling the processing of higher-speed signals and reducing the impact of photoelectric conversion rate on optical computing speed. Therefore, the computing speed of the neuron can be improved. Simultaneously, UTC-PD can operate under zero bias conditions, consuming less energy than other schemes, thus reducing the energy consumption of the neuron.

[0034] Alternatively, in one embodiment of this disclosure, Figure 2 This is a schematic diagram illustrating the change of photocurrent with light intensity, provided in an embodiment of this disclosure. Figure 3 This is a schematic diagram illustrating the variation of photocurrent difference with light intensity, provided in an embodiment of this disclosure. Figure 2 and Figure 3 As shown, the first saturation light intensity P2 corresponding to the first single-row carrier photodetector PD1 is greater than the second saturation light intensity P1 corresponding to the second single-row carrier photodetector PD2. The correspondence between the photocurrent difference (I1-I2) and the light intensity includes: When the light intensity is less than the second saturation light intensity P1, both the first single-row carrier photodetector PD1 and the second single-row carrier photodetector PD2 operate in the linear region. Furthermore, the first responsivity of the first single-row carrier photodetector PD1 and the second responsivity of the second single-row carrier photodetector PD2 are the same; that is, the slope of the photoresponse curve corresponding to the first single-row carrier photodetector PD1 is the same as the slope of the photoresponse curve corresponding to the second single-row carrier photodetector PD2. Therefore, the first photocurrent I1 is the same as the second photocurrent I2. Figure 2 As shown. At this time, the difference in photocurrent (I1-I2) is zero, as... Figure 3 As shown.

[0035] When the light intensity is not less than the second saturation light intensity P1 and less than the first saturation light intensity P2, the first single-row carrier photodetector PD1 still operates in the linear region, while the second single-row carrier photodetector PD2 enters the saturation state. At this time, the first responsivity remains unchanged, the second responsivity is zero, the second photocurrent I2 maintains the second saturation photocurrent unchanged, and the first photocurrent I1 continues to increase proportionally to the light intensity. Figure 2 As shown. At this time, the difference in photocurrent (I1-I2) is proportional to the light intensity, and its slope is the same as the slope of the photoelectric response curve corresponding to the first single-row carrier photodetector PD1, as shown. Figure 3 As shown.

[0036] When the light intensity is not less than the first saturation light intensity P2, both the first single-row carrier photodetector PD1 and the second single-row carrier photodetector PD2 enter the saturation state. At this time, both the first and second responsivity are zero, the first photocurrent I1 maintains the first saturation photocurrent unchanged, and the second photocurrent I2 maintains the second saturation photocurrent unchanged. Figure 2 As shown. At this time, the difference in photocurrent (I1-I2) is a constant value, as... Figure 3 As shown.

[0037] It should be noted that when the light intensity is less than the second saturation light intensity P1, the first responsivity corresponding to the first single-row carrier photodetector PD1 and the second responsivity corresponding to the second single-row carrier photodetector PD2 are the same, but neither is zero. When the light intensity is not less than the second saturation light intensity P1 and less than the first saturation light intensity P2, the first responsivity remains unchanged at the value when the light intensity is less than the second saturation light intensity P1.

[0038] In some embodiments, optical power is the light intensity input to the first single-row carrier photodetector PD1 and the second single-row carrier photodetector PD2.

[0039] According to some embodiments, the absorption layer thickness of the first single-row carrier photodetector PD1 and the second single-row carrier photodetector PD2 is the same, so that when the light intensity is less than the first saturation light intensity P1, the first responsivity of the first single-row carrier photodetector and the second responsivity of the second single-row carrier photodetector are the same. The first single-row carrier photodetector and the second single-row carrier photodetector have different structural dimensions or other epitaxial layer doping concentrations, except for the absorption layer thickness, so that the first saturated photocurrent and the second saturated photocurrent are different. Specifically, the first saturated photocurrent is greater than the second saturated photocurrent.

[0040] In some embodiments, other structural dimensions may include, but are not limited to, the thickness and doping concentration of the cliff layer, the thickness and doping concentration of the collection layer, the ratio of the thickness of the depletion absorber layer to the thickness of the non-depletion absorber layer, and the doping concentration.

[0041] Optionally, in one embodiment of this disclosure, the neuron based on a single-row carrier photodetector further includes a light source, a linear computing unit, and an output unit; wherein, The light source is connected to the input terminal of the linear computing unit and is used to output at least two first light signals to the linear computing unit; The output of the linear calculation unit is connected to the input of the nonlinear activation function unit, and is used to perform linear calculations on at least two first light signals input from the light source to obtain and output a second light signal to the nonlinear activation function unit. The output of the nonlinear activation function unit is connected to the output unit and is used to divide the second optical signal into at least two third optical signals, and convert the third optical signals into current signals according to the single-row carrier photodetector. The output unit is used to convert the current difference signal between at least one pair of single-row carrier photodetectors into a voltage signal and output the voltage signal.

[0042] According to some embodiments, the light source includes at least two semiconductor lasers, and the semiconductor lasers include at least one of the following: Distributed feedback semiconductor laser; Distributed Bragg semiconductor laser; Vertical cavity surface-emitting semiconductor laser.

[0043] In some embodiments, the wavelengths of the first optical signals output by each of the at least two semiconductor lasers are different; for example, the wavelengths of the first optical signals output by any two adjacent semiconductor lasers may be spaced several nanometers apart.

[0044] According to some embodiments, the linear computing unit includes at least two first amplitude modulators, at least two second amplitude modulators, a wavelength division multiplexer (WDM), and a semiconductor laser, with each of the first and second amplitude modulators corresponding to one another; wherein... The input terminal of the first amplitude modulator is connected to the output terminal of the semiconductor laser, and is used to perform amplitude modulation on the first optical signal according to the input signal to obtain the first sub-optical signal; The input of the second amplitude modulator is connected to the output of the first amplitude modulator, and is used to perform amplitude modulation on the first sub-optical signal according to the weight signal to obtain the second sub-optical signal. The input of the wavelength division multiplexer (WDM) is connected to the output of at least two second amplitude modulators to perform multiplexing of at least two second sub-optical signals to obtain a second optical signal.

[0045] In some embodiments, the amplitude modulator used in this disclosure, such as a first amplitude modulator and a second amplitude modulator, each includes at least one of the following: Electroabsorption modulator; Mach-Zehnder modulator; Micro-ring modulator.

[0046] In some embodiments, the amplitude modulator includes a beam splitter that can split the first optical signal input from the semiconductor laser into at least two paths. The beam splitter can be composed of a multimode interferometer, a directional coupler, or a cascaded Y-branch.

[0047] It should be noted that the amplitude modulator array (including at least two first amplitude modulators and at least two second amplitude modulators) in the linear computing unit can realize linear multiplication and addition operations. The first optical signal of each wavelength is multiplied after passing through the two amplitude modulators. Then, the second sub-optical signals of different wavelengths are combined by the wavelength division multiplexer (WDM) to complete the addition function.

[0048] In some embodiments, wavelength division multiplexers (WDMs) include, but are not limited to, arrayed waveguide gratings, Mach-Zehnder type wavelength division multiplexers, etc.

[0049] To give an example from a scenario, Figure 4 This is a schematic diagram of the neuron structure based on a single-row carrier photodetector provided in an embodiment of this disclosure. Figure 4 As shown, the light source includes four semiconductor lasers, namely LD1, LD2, LD3, and LD4. The linear computing unit includes four first amplitude modulators, namely Mod1, Mod2, Mod3, and Mod4, and four second amplitude modulators, namely Mod5, Mod6, Mod7, and Mod8. Connect LD1, Mod1, and Mod5 in sequence; connect LD2, Mod2, and Mod6 in sequence; connect LD3, Mod3, and Mod7 in sequence; and connect LD4, Mod4, and Mod8 in sequence. The wavelength division multiplexer (WDM) is connected to the outputs of Mod5, Mod6, Mod7, and Mod8 respectively, and can combine four second sub-optical signals of different wavelengths into the same waveguide to achieve the summation function.

[0050] According to some embodiments, the nonlinear activation function unit includes a multimode interferometer; wherein... The input of the multimode interferometer is connected to the output of the linear computing unit to perform equal-splitting processing on the second optical signal to obtain at least two third optical signals; The input terminal of the single-carrier photodetector is connected to the output terminal of the multimode interferometer to convert the third optical signal into a current signal; The output terminal of the single-row carrier photodetector is connected to the output unit.

[0051] In some embodiments, such as Figure 4 As shown, the nonlinear activation function unit includes only one pair of single-row carrier photodetectors, namely PD1 and PD2, and the multimode interferometer is a 1×2 multimode interferometer (MMI); among which, The input of the 1×2 multimode interferometer (MMI) is connected to the output of the wavelength division multiplexer (WDM) to split the combined second optical signal into two third optical signals. The output of the 1×2 multimode interferometer (MMI) is connected to the inputs of PD1 and PD2, respectively.

[0052] According to some embodiments, the output unit includes a transimpedance amplifier (TIA), wherein, The input of the transimpedance amplifier TIA is connected to the output of the nonlinear activation function unit.

[0053] In some embodiments, such as Figure 4 As shown, the input terminal of the transimpedance amplifier TIA is connected to the output terminals of PD1 and PD2 respectively, which can convert the current difference signal between PD1 and PD2 into a voltage signal and output the voltage signal.

[0054] It should be noted that the InP-based UTC-PD can be monolithically integrated with semiconductor lasers and modulators, simultaneously implementing both linear weighting and nonlinear activation functions. Therefore, the neuron based on a single-row carrier photodetector provided in this disclosure can be integrated onto a single chip to simultaneously implement both linear weighting and nonlinear activation functions. Furthermore, the UTC-PD can also serve as a monitor to detect the results of linear calculations in the optical neural network.

[0055] In summary, the InP optoelectronic chip integrating a neuron based on a single-row carrier photodetector provided in this embodiment includes a nonlinear activation function unit. The nonlinear activation function unit includes at least one pair of single-row carrier photodetectors, and any one pair of single-row carrier photodetectors includes a first single-row carrier photodetector and a second single-row carrier photodetector. The first saturated photocurrent corresponding to the first single-row carrier photodetector and the second saturated photocurrent corresponding to the second single-row carrier photodetector are different. The nonlinear activation function unit is used to implement the activation function based on the correspondence between the difference in photocurrent and the light intensity, where the difference in photocurrent is the difference between the first photocurrent output by the first single-row carrier photodetector and the second photocurrent output by the second single-row carrier photodetector. Therefore, by combining the saturated absorption effect of the single-row carrier photodetector in the nonlinear activation function unit, a higher saturated output power and a larger bandwidth can be achieved, enabling the processing of higher-speed signals and reducing the impact of the photoelectric conversion rate on the optical computing speed, thereby improving the computing speed of the neuron. Meanwhile, single-carrier photodetectors can operate under zero bias conditions, resulting in lower energy consumption compared to other solutions, which can reduce the energy consumption of neurons.

[0056] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms may refer to different embodiments or examples. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0057] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0058] Although embodiments of this disclosure have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of this disclosure, the scope of which is defined by the claims and their equivalents.

Claims

1. An InP optoelectronic chip integrating neurons based on a single-row carrier photodetector, characterized in that, The neuron based on a single-row carrier photodetector includes: a nonlinear activation function unit, wherein the nonlinear activation function unit includes at least one pair of single-row carrier photodetectors, and any one pair of single-row carrier photodetectors includes a first single-row carrier photodetector and a second single-row carrier photodetector; wherein... The first saturated photocurrent corresponding to the first single-row carrier photodetector and the second saturated photocurrent corresponding to the second single-row carrier photodetector are different. The nonlinear activation function unit is used to realize the function of activation function according to the correspondence between the difference of photocurrent and the light intensity. The difference of photocurrent is the difference between the first photocurrent output by the first single-row carrier photodetector and the second photocurrent output by the second single-row carrier photodetector. Wherein, the first saturated light intensity corresponding to the first single-row carrier photodetector is greater than the second saturated light intensity corresponding to the second single-row carrier photodetector, and the correspondence between the difference in photocurrent and the light intensity includes: When the light intensity is less than the second saturation light intensity, the first responsivity corresponding to the first single-row carrier photodetector and the second responsivity corresponding to the second single-row carrier photodetector are the same, and the difference in photocurrent is zero. When the light intensity is not less than the second saturation light intensity and less than the first saturation light intensity, the first responsivity remains unchanged, the second responsivity is zero, the second photocurrent maintains the second saturation photocurrent unchanged, and the difference in photocurrent is proportional to the light intensity. When the light intensity is not less than the first saturation light intensity, both the first responsivity and the second responsivity are zero, the first photocurrent remains unchanged at the first saturation photocurrent, the second photocurrent remains unchanged at the second saturation photocurrent, and the difference between the photocurrents is a constant value.

2. The InP optoelectronic chip integrating neurons based on a single-row carrier photodetector according to claim 1, characterized in that, The first single-row carrier photodetector and the second single-row carrier photodetector have the same absorption layer thickness, so that when the light intensity is less than the second saturation light intensity, the first responsivity of the first single-row carrier photodetector and the second responsivity of the second single-row carrier photodetector are the same. The first single-row carrier photodetector and the second single-row carrier photodetector have different structural dimensions or other epitaxial layer doping concentrations, except for the absorption layer thickness, so that the first saturated photocurrent and the second saturated photocurrent are different.

3. The InP optoelectronic chip integrating neurons based on a single-row carrier photodetector according to claim 2, characterized in that, The other structural dimensions include the thickness and doping concentration of the cliff layer, the thickness and doping concentration of the collection layer, the ratio of the thickness of the depletion absorber layer to the thickness of the non-depletion absorber layer, and the doping concentration.

4. The InP optoelectronic chip integrating neurons based on a single-row carrier photodetector according to claim 1, characterized in that, The neuron based on a single-row carrier photodetector further includes a light source, a linear computation unit, and an output unit; wherein... The light source is connected to the input terminal of the linear computing unit and is used to output at least two first light signals to the linear computing unit. The output of the linear calculation unit is connected to the input of the nonlinear activation function unit, and is used to perform linear calculations on at least two first light signals input from the light source to obtain and output a second light signal to the nonlinear activation function unit. The output terminal of the nonlinear activation function unit is connected to the output unit, and is used to divide the second optical signal into at least two third optical signals, and convert the third optical signals into current signals according to the single-row carrier photodetector. The output unit is used to convert the current difference signal between the at least one pair of single-row carrier photodetectors into a voltage signal and output the voltage signal.

5. The InP optoelectronic chip integrating neurons based on a single-row carrier photodetector according to claim 4, characterized in that, The light source includes at least two semiconductor lasers, and the semiconductor lasers include at least one of the following: Distributed feedback semiconductor laser; Distributed Bragg semiconductor laser; Vertical cavity surface-emitting semiconductor laser.

6. The InP optoelectronic chip integrating neurons based on a single-row carrier photodetector according to claim 5, characterized in that, The linear computing unit includes at least two first amplitude modulators, at least two second amplitude modulators, and a wavelength division multiplexer, wherein the semiconductor laser, the first amplitude modulator, and the second amplitude modulator correspond one-to-one; wherein... The input terminal of the first amplitude modulator is connected to the output terminal of the semiconductor laser, and is used to perform amplitude modulation on the first optical signal according to the input signal to obtain the first sub-optical signal; The input terminal of the second amplitude modulator is connected to the output terminal of the first amplitude modulator, and is used to perform amplitude modulation on the first sub-optical signal according to the weight signal to obtain the second sub-optical signal; The input terminal of the wavelength division multiplexer is connected to the output terminals of the at least two second amplitude modulators, and is used to perform multiplexing processing on the at least two second sub-optical signals to obtain the second optical signal.

7. The InP optoelectronic chip integrating neurons based on a single-row carrier photodetector according to claim 6, characterized in that, Both the first amplitude modulator and the second amplitude modulator include at least one of the following: Electroabsorption modulator; Mach-Zehnder modulator; Micro-ring modulator.

8. The InP optoelectronic chip integrating neurons based on a single-row carrier photodetector according to claim 4, characterized in that, The nonlinear activation function unit includes a multimode interferometer; wherein... The input end of the multimode interferometer is connected to the output end of the linear computing unit, and is used to perform equal-division processing on the second optical signal to obtain at least two third optical signals; The input terminal of the single-row carrier photodetector is connected to the output terminal of the multimode interferometer, and is used to convert the third optical signal into a current signal; The output terminal of the single-row carrier photodetector is connected to the output unit.

9. The InP optoelectronic chip integrating neurons based on a single-row carrier photodetector according to claim 4, characterized in that, The output unit includes a transimpedance amplifier, wherein, The input terminal of the transimpedance amplifier is connected to the output terminal of the nonlinear activation function unit.

Images