Apparatus, system, and method for providing artificial neuron networks

Abstract

The invention relates to an apparatus (10) for providing an artificial neuron network (200), having: - at least one optical neuron component (220) for providing a neuron (251) of the network (200), wherein the neuron component (220) has a microresonator (240) in order to output an output signal (222) of the neuron component (220) in a non-linearly dependent manner on an input signal (221) of the neuron component (220).

Description

Technical Field

[0001] This invention relates to an apparatus for providing an artificial neural network. Furthermore, this invention relates to systems and methods for this purpose. Background Technology

[0002] The use of artificial neural networks in an increasing number of applications is known from existing technologies. For example, in the field of vehicle technology, neural networks can significantly improve the reliability of autonomous driving functions, such as driver assistance systems.

[0003] The advantages of using neural networks are described below, exemplarily based on such vehicle functionality. It is known that reliable environmental detection is crucial for autonomous driving. Here, the vehicle's environment is detected using sensors such as radar, lidar, and camera sensors. Furthermore, a holistic 360°-3D detection of the environment can be used, enabling the detection and classification of all static and dynamic objects.

[0004] For example, this neural network can classify camera images from the vehicle's front-facing camera. Here, the neural network can associate categories with individual pixels in the camera image (e.g., different categories for roads, road markings, vehicles, pedestrians, and / or vegetation in the environment). This category information allows for more accurate environmental detection. Where possible, pixel-by-pixel association of the environment can be performed. Furthermore, this information aids in scene understanding, enabling vehicle functions to adapt accordingly.

[0005] In particular, cameras play a crucial role in redundant and robust environment detection because this type of sensor can accurately measure angles for environment detection and can be used for environment classification. However, the processing and classification of camera images is computationally intensive and structurally complex. In particular, a problem in 360°-3D environment detection is that many individual images must be classified and processed, thereby increasing computational costs.

[0006] Conventional high-performance artificial neural networks (NNs or KNNs) have already provided the possibility of classifying camera images or other sensor data with image refresh rates below 10Hz. Therefore, processing and classification can be significantly accelerated.

[0007] However, in some cases, this is still insufficient or can be improved, as modern camera systems operate at an image refresh rate of 30Hz. For example, vehicle functions may require reliable real-time environmental detection, necessitating processing and classification with minimal time consumption.

[0008] Furthermore, the data load increases with the resolution capabilities of camera images. For example, modern car cameras already offer 8 megapixel resolution. Currently, classifying these camera images in a vehicle in real time is either impossible or technically very complex. This limitation, particularly processor speed, even when using a GPU (Graphics Processing Unit), is insufficient for comprehensive real-time image classification and processing.

[0009] Classification is particularly necessary for understanding the context of the environment in order to handle driving maneuvers in accordance with the vehicle's surroundings. Therefore, incomplete or incorrect classification is a problem for autonomous driving functions and driver assistance systems.

[0010] In summary, the problem is that conventional neural networks are limited by processor speed in terms of computational power, and therefore cannot provide sufficient capability for certain applications. High-resolution camera images and sensor data can be classified using neural networks, for example, only at reduced image refresh rates below 10Hz.

[0011] The possibility of using artificial neural networks in a manner similar to that used with optical components is known from document WO 2017 / 210550 A1. Furthermore, a conventional method is known from WO 2008 / 136886 A1. Summary of the Invention

[0012] The objective of this invention is to at least partially eliminate the aforementioned drawbacks. In particular, the objective of this invention is to achieve an increase in the computational speed of neural networks, and especially to enable classification and processing to be performed in real time and thus with a predetermined time constraint via neural networks.

[0013] The aforementioned tasks are solved by means of the apparatus according to the invention, the system according to the invention, and the method according to the invention. Further features and details of the invention are derived from the specification and drawings. The features and details described in conjunction with the apparatus according to the invention apply, as do the system according to the invention and the method according to the invention, and vice versa, so that the disclosures regarding the various aspects of the invention are always mutually referential or can be mutually referenced.

[0014] This task is particularly solved by an apparatus for providing an artificial neural network (KNN), preferably an optical neural network, having at least one optical neuron component (and particularly a large number of optical neuron components, respectively) for providing neurons for the network.

[0015] Specifically, the corresponding neuronal component is configured to have a microresonator to output the neuronal component's output signal in a manner nonlinearly related to the input signal of the neuronal component. In other words, the input signal can be nonlinearly converted into an output signal. Accordingly, the present invention can achieve optical processing of the neuronal network at the speed of light by using at least one optical microresonator, particularly a microring resonator (MRR for short), as the corresponding neuron and, if possible, additionally using optical weights. This provides the advantage of higher processing speed and, if possible, a more compact network structure compared to conventional optical neuronal networks. To obtain the input signal, input information can be received, for example, from an interface component. If the input information already exists in optical form, such as light from the optical device of a sensor or lidar radiation, its optics can advantageously be directly used as the input signal for one or more neuronal components. Otherwise, the input information can be pre-converted into an optical input signal. The input information or input signal can, for example, be transmitted to an optical carrier, and the nonlinear response function can be generated in the optical carrier by the microresonator, for example, via a four-wave mixing process.

[0016] It is possible that the microresonator provides a nonlinear response function for the neural network. The output signal can be derived from this nonlinear response function. Depending on the possible network configuration of KNN, the output signal can be forwarded to the next neuron unit via at least one optical weight, i.e., forwarded to the microresonator of the neuron unit in the next neuron layer and / or the neuron unit in that next neuron layer.

[0017] For example, the network and its corresponding output signal can be evaluated based on the detection of intensity and / or phase and / or frequency. It is also conceivable that the optical output signal is converted into electrical output information by an interface component for further processing by a processor such as a CPU or GPU. In this way, neural networks, especially deep optical neural networks, can be provided particularly efficiently. Due to the use of microresonators, the optical power can be kept very low, thus resulting in lower power requirements for the light source of the input signal (such as a laser). It is possible to mount a large number of microresonators for the network onto a carrier, such as a chip, with minimal space consumption. Therefore, optical networks can be implemented on the chip surface. The entire optical neural network can be physically fabricated on a semiconductor and, for example, integrated into a semiconductor chip.

[0018] The corresponding microresonator can be, for example, a microring resonator, preferably used to provide optical nonlinear effects such as self-phase modulation in order to provide neurons in an artificial neural network. Specifically, the microresonator can be implemented to provide nonlinear functions of neurons, preferably activation functions such as sigmoid functions. The device according to the invention can be implemented as an optical and / or optoelectronic device for providing artificial neural networks (hereinafter also simply referred to as KNNs or NNs), especially optical KNNs.

[0019] Furthermore, within the scope of this invention, the microresonator can be configured as a microring resonator. For example, a ring wave conductor can be configured to construct a microring resonator. Additionally, the microring resonator can be implemented to provide optical nonlinear effects to generate an output signal and / or provide nonlinear functionality of a neuron.

[0020] Optionally, the neuronal components can be configured such that each component has at least one or exactly one additional microresonator, wherein the microresonators are arranged adjacently (particularly at a predetermined distance) to construct exactly one neuron. For example, the function and / or weight of the neuron can be adjusted by matching the distance and / or diameter of the microresonators. Furthermore, the microresonators can be configured to have different diameters.

[0021] Another advantage within the scope of this invention can be obtained by implementing the microresonator to provide optical nonlinear effects in order to define the nonlinear relationship between the input and output signals. For example, multiphoton processes of the microresonator can be used for this purpose.

[0022] Furthermore, it is advantageous that the neuron component is configured to nonlinearly transform the input signal of the neuron component by means of at least one optical nonlinear effect to generate the output signal of the neuron component, and / or to provide activation function with the input signal of the neuron component as input by means of at least one optical nonlinear effect. In the device according to the invention, the neuron component can be correspondingly configured to be implemented as a nonlinear optical component. Here, depending on the input signal of the neuron component having a first frequency, the output signal of the neuron component can be output at a second frequency, wherein the second frequency is different from the first frequency. In other words, the neuron component can be implemented to perform frequency conversion of the input signal to obtain the output signal. The input signal and the output signal can be implemented here as optical signals, i.e., light or light beams and / or laser beams, respectively. This enables the optical provision of the neuron function of the KNN. Due to the nonlinear implementation of the neuron component, the frequency of the output signal may be nonlinearly related to the input signal (e.g., related to parameters of the input signal, such as frequency and / or amplitude and / or phase and / or polarization). This nonlinear correlation enables the provision of the neuron function, such as activation function. Nonlinear mapping can be implemented by the neuron component, for example, in the form of a sigmoid function. In other words, the difference between the second frequency and the first frequency, or the frequency transition, can be defined by the nonlinear correlation between the output signal and the input signal. Due to nonlinear effects, an increase in the aforementioned parameters of the input signal (such as the frequency or wavelength of light) can lead to an increase in the frequency or wavelength of the output signal according to the S-shape of the S-type. The nonlinear mapping of the neuronal component can be shown, for example, in the form y=f(x), and in the case of an S-shaped function, as y=sig(x). The parameters x and y here can represent the frequency or wavelength of light, respectively, which can therefore be used as the input and output signals.

[0023] In particular, the present invention is based on the idea of ​​using optical neurons and optical weights for optical processing of data to improve the computational speed of KNN. This has the advantage that, in principle, data can be processed at the speed of light. Here, the data corresponds to input information, i.e., the input to the KNN, such as image information, like camera images. The input information can be received electronically, for example, by a device, but then converted into optical information to obtain optical input signals for the neuronal components. Since the NN (typically) has multiple neurons, multiple input signals can be formed from the input information for multiple neuronal components accordingly. Furthermore, the input information and / or the optical information thus obtained can be pre-processed and / or weighted, if possible, to obtain at least one input signal. In particular, weighting by weighting components will be discussed in more detail below.

[0024] Furthermore, it is possible to configure the neuronal components to provide at least one weight for the network so as to output an output signal based on a weighted sum (by the weights) of the input signal. Alternatively or additionally, it is conceivable that the weights are configured to linearly transform the input signal to produce the output signal, wherein the weights are defined at least in part by the arrangement and / or configuration of the microresonators. It is also conceivable that each neuronal component has at least one or exactly one additional microresonator, wherein the microresonators are arranged adjacent to each other, and the distance between the microresonators and / or the diameter of the microresonators defines the weights.

[0025] Optical materials used for optical processing can be used for neurons and / or weights in a KNN, thus enabling the KNN to be implemented as an optical KNN. Accordingly, at least one neuronal component or microresonator can have optical materials to provide a nonlinear process. In this way, the nonlinear process of the neuronal component or microresonator can implement the function of the neuron. Thus, the input signal or its information is processed at almost the speed of light. Furthermore, waveguides or similar microresonators can be used as weights in the KNN. Special optical materials can be used to match the properties of the waveguides or microresonators, thereby enabling weighting in an additive, subtractive, or multiplicative manner.

[0026] It is conceivable that the device according to the invention further includes: - At least one (or more) optical weighting components (each) are used to provide at least one weight of the network so as to output a weighting component output signal based on the weighting of the input signal of the weighting component, wherein the weighting component preferably performs the weighting of the input signal optically.

[0027] Weighting components can be used to provide at least one weight, for example, by having a transparent material and / or a doped material and / or a material with defined absorption of the transmitted light and / or an optical amplifier or attenuator. It is also conceivable that microresonators have weighting components. The input signal to the weighting element is, for example, light incident on the material, which passes through the medium of the weighting component and is then emitted, so that the output signal of the weighting component can be the emitted light.

[0028] Furthermore, the neuron component can be optically connected to the weighting component, for example via optical or waveguides, to at least partially form the input signal of the neuron component from the output signals of the weighting component and, if possible, other weighting components. In this way, the weighting component can be used to change the weights of the neuron. Thus, the classic structure of a KNN can be constructed optically using the weighting component and the neuron component. One possible topology of a KNN is a recurrent neural network or a single-layer or multi-layer feedforward network. Similarly, the construction of a convolutional neural network (CNN) is also conceivable optically according to the present invention.

[0029] Furthermore, the connection can also occur in such a way that the output signal of the weighting component corresponds to the input signal of the neuron component to which the weighting component is associated. Additionally, the weighting component can be fixedly associated with the neuron component to perform weighting of the neuron's input. This can be advantageously achieved through a fixed optical connection between the weighting component and the neuron component.

[0030] Similarly, the present invention relates to a system for vehicles, particularly motor vehicles and / or autonomous vehicles, having: - The device according to the invention, - At least one vehicle component.

[0031] Therefore, the system according to the invention brings the same advantages as those described in detail with reference to the device according to the invention.

[0032] Alternatively, the device according to the invention may have electronic and / or optoelectronic interface components so as to: - Receive electrical input information from vehicle components - Based on the received input information, provide optical input signals for neuronal components or multiple optical input signals for a large number of neuronal components to the neural network. - Provide electrical output information to vehicle components based on the optical output signals of the neuron components.

[0033] Another advantage that can be obtained within the scope of the invention is that at least one vehicle component has a detection device (such as a camera) to generate input information in the form of image information, and wherein at least one vehicle component has a driver assistance system for providing autonomous driving functions, so as to evaluate output information by the driver assistance system and use the output information as a classification of the vehicle's environment.

[0034] The KNN provided by the present invention can be implemented to perform real-time classification and processing of input, particularly image information, as input information with a predetermined finite time consumption. This can be achieved by constructing the KNN at least partially from optical components that perform optical processing. In particular, the activation function of the neurons in the KNN can be implemented optically. Therefore, the KNN can be implemented optically. Here, the frequency of the input and / or output signals can be used as a parameter to be processed for the activation function. Thus, in the electronic implementation of the KNN, frequency forms a counterpart to voltage.

[0035] Similarly, the object of the present invention is a method for providing an artificial neural network. Specifically, it is configured to perform the following steps, preferably one after another or in any order, wherein these steps may also be repeated if possible: - The neurons of the network are provided by at least one optical neuron component, wherein the neuron component (for this purpose) has a microresonator. - The output signal of the neuron component is output by the microresonator in a manner that is nonlinearly related to the input signal of the neuron component.

[0036] Therefore, the method according to the invention brings the same advantages as those described in detail with reference to the apparatus according to the invention. Furthermore, the method can be applied to operating the apparatus according to the invention. It is also advantageous to implement these steps using the apparatus or system according to the invention.

[0037] In another possibility, the nonlinear correlation is provided by nonlinearly transforming the input signal using a microresonator to generate an output signal, wherein self-phase modulation of the microresonator is used for this purpose. It is possible here that all parts of the artificial neural network are optically mapped using the device according to the invention. For example, neurons in a KNN can be provided separately by neuronal components or microresonators, and / or weighting can be provided by weighting components or microresonators.

[0038] Another advantage is that the amplitude or frequency of the input signal can be non-linearly altered by the neuron to obtain the output signal. It is possible that weighting can also be provided by changing the frequency and / or amplitude. Here, the function and / or weighting can be adjusted by changing the structure of the microresonator. For example, changes in amplitude can be influenced by the absorption characteristics of the microresonator, or changes in frequency can be influenced by adjusting the diameter of the microresonator.

[0039] The device according to the invention may have at least one or preferably multiple optical neuron components (each) for providing at least one neuron of the network. The neuron components may be implemented to optically provide the function of a neuron in a KNN. This has the advantage that the KNN at least partially performs optical processing, and therefore can perform processing at a higher speed than a conventional electronic KNN. Unlike conventional optical methods in KNNs, the device according to the invention is particularly configured to use optical nonlinear effects (i.e., the effects of nonlinear optical devices) to provide the neuron's function and, in particular, its activation function.

[0040] Furthermore, it is conceivable that the neuron components are configured to output a neuron component with a second amplitude and / or phase, depending on the input signal of the neuron component having a first amplitude and / or phase, wherein the second amplitude and / or phase differs from the first amplitude and / or phase. Accordingly, it is possible that, instead of (only) optically processing the frequency as a parameter of the input and / or output signals, other parameters such as amplitude and / or phase are also optically processed. This could further improve the reliability of the processing.

[0041] Within the scope of this invention, the electronic and / or optoelectronic interface components, particularly for at least one electronic vehicle component, can be advantageously configured to provide an artificial neural network (KNN) within the vehicle. The interface component can convert electrical input information (e.g., in the form of digital data and / or electrical signals) into optical information, enabling processing via an optical KNN. Subsequently, electrical output information can be formed from the optical output signal of the neural component through the interface component or another interface component. Here, contrary to the described input and output signals, the electrical information is not transmitted optically but through electrical conductors. Accordingly, the device according to the invention can be configured to connect to an electronic device, particularly a vehicle component, via a cable through the interface component for electrical transmission of input and output information.

[0042] Furthermore, it is advantageous that the vehicle is constructed as a motor vehicle, particularly a land motor vehicle without tracks. Thus, the vehicle can be constructed, for example, as a hybrid vehicle comprising an internal combustion engine and a traction motor, or as a (pure) electric vehicle, or constructed with only an internal combustion engine. The vehicle can preferably be implemented with a high-voltage onboard network and / or an electric motor. The vehicle can also be constructed as a fuel cell vehicle. Additionally, the vehicle can also be a passenger car or a truck. In the implementation of an electric vehicle, it is preferable that no internal combustion engine is installed in the vehicle, and it is then driven solely by electrical energy.

[0043] Furthermore, it is conceivable that at least one vehicle component has a detection device, such as a camera, to generate input information in the form of image information, and / or that at least one vehicle component has a driver assistance device for providing autonomous driving functions, preferably evaluating output information through the driver assistance system, and using the output information here as a classification of the vehicle's environment. Here, using an optical KNN can achieve real-time evaluation of complete image information. The detection device includes, for example, radar and / or lidar and / or ultrasound, or at least one radar and / or lidar and / or ultrasound sensor, and / or at least one camera, especially a front-facing camera for the vehicle. The detection device can be implemented for detecting the vehicle's environment, particularly in the direction of travel. Attached Figure Description

[0044] Other advantages, features, and details of the invention will become apparent from the following description, in which embodiments of the invention are described in detail with reference to the accompanying drawings. Herein, the features mentioned in the claims and specification may be important to the invention individually or in any combination. Wherein: Figure 1 A schematic diagram of a neural network is shown. Figure 2 Schematic diagrams of the apparatus and system according to the invention are shown. Figure 3 A schematic diagram for visualizing the method according to the present invention is shown. Figures 4 to 8 Possible embodiments of the components of the device according to the invention are shown.

[0045] In the following figures, the same reference numerals are used for the same technical features even in different embodiments. Detailed Implementation

[0046] Neural networks are required for many applications. For example, in vehicle 8, data processing can benefit from the classification of images of the environment detected by the neural network 200 used for driver assistance system 7. Thus, the safest possible environmental perception is crucial for autonomous driving. Here, the environment is detected by means of detection devices 6 of vehicle 8, such as radar, lidar, and cameras. 360° three-dimensional detection of the environment is particularly advantageous here, allowing for the detection and classification of relevant static and dynamic objects. In particular, cameras play a key role in redundant and robust environment detection because this type of sensor can accurately measure angles in environment detection and can be used to classify the environment. However, the processing and classification of camera images is computationally intensive and structurally complex. In particular, a problem in 360° environment detection is that many individual images must be classified and processed, thereby increasing computational costs. The device 10 according to the invention can help solve this problem.

[0047] Artificial neural network 200 can be used for autonomous driving functions that classify the environment. Here, sensor data can be electronically transmitted to neuron 251 via weighting 252. The output signal 222 of neuron 251, and thus the transmitted signal to the neurons 251 of the following layer, is preferably given here by a sigmoid function of the sum of the weighted response functions. in The weight is 252. It is neuron 251 Functions. Artificial neural networks 200 thus form functions. in And the function value Output as category information.

[0048] exist Figure 1 In this context, such an artificial neural network 200 is schematically shown as, for example, having input and output information 231, 232 and neurons 251 and weights 252 formed by neuronal components 220, respectively.

[0049] exist Figure 2The image schematically illustrates a device 10 according to the invention for providing an artificial neural network 200. The device 10 according to the invention has at least one optical neuron component 220 for providing neurons 251 of the network 200. Here, the neuron component 220 is configured to have (in...) Figures 4 to 8 (As shown in more detail below) microresonator 240, so as to output the output signal 222 of neuron component 220 in a manner nonlinearly related to the input signal 221 of neuron component 220.

[0050] The device 10 according to the invention can preferably be configured as part of the system 1 for a vehicle 8 according to the invention, as in the same way. Figure 2 As shown in the diagram. Here, neural network 200 is used for at least one vehicle component 5, which may be, for example, the driver assistance system 7 described above. Accordingly, neural network 200 can be used to classify images from detection device 6, for example, distinguishing between static and dynamic objects in the images. Images may be provided to at least one neural component 220, for example, in the form of input information 231 and via interface component 20. For this purpose, if the input information 231 should not already be optical, it may optionally be converted into an optical input signal 221.

[0051] Method steps 101-102 of the method according to the present invention are as follows: Figure 3 The diagram is schematically shown. According to the first step 101, the neurons 251 of the network 200 are provided by at least one optical neuron component 220, wherein the neuron component 220 has a microresonator 240. According to the second step 102, the output signal 222 of the neuron component 220 is at least partially generated by the microresonator 240 in a manner nonlinearly related to the input signal 221 of the neuron component 220.

[0052] The micro-resonator 240, in the form of a Mikro-Ring resonator, is... Figure 4The diagram is schematically shown. The input signal 221 is coupled into the waveguide 210, for example, in the form of a continuous wave (CW) laser beam, although a pulsed laser could also be used in principle. The input signal 221 could also originate from input information 231 and be coupled via the optical interface assembly 20. The waveguide 210 can be constructed as part of an optical semiconductor. Subsequently, in the interaction region 211, the coupled input signal 221 can be coupled as an evanescent field into the microresonator 240 (also called a resonator ring), particularly the ring waveguide 243 (abbreviated as: ring). The ring waveguide 243 can be arranged as a ring waveguide structure, at a very small distance A from the waveguide 210 on the semiconductor, which is in the form of a linear waveguide structure. If the distance A between the two waveguides 210, 243 is so small that the evanescent field of electromagnetic radiation in the linear waveguide 210 extends into the ring waveguide 243, and the radiation from the linear waveguide 210 is coupled into the ring waveguide 243, where it propagates. The optical path length of the ring wave conductor 243 can be chosen such that it is an integer multiple of the wavelength. In this way, light propagating in the ring wave conductor 243 can constructively interfere with the coupled evanescent field after a cycle and produce enhancement. If the interaction region 211 between the linear and ring wave conductors 210 is within the wavelength range, the interaction between the two fields has only a short duration, resulting in constructive interference. Therefore, the neuron component 220 shown forms an optical resonator.

[0053] After a period, the newly coupled field and the existing field in the loop can constructively interfere. In this way, amplitude modulation can be generated. Conversely, a small portion can be decoupled. Due to the increased intensity in the loop, a basic soliton can be constructed, which is decoupled as an output signal 222 in the form of a pulse train.

[0054] A portion of the light propagating within the ring waveguide 243 is decoupled back into the linear waveguide 210 after each cycle and can be used as a signal. Through the microring resonator, the light can be amplitude modulated by appropriately selecting the diameter D and the coupling ratio of the waveguide 210, thereby forming an output signal 222 from the input signal 221, particularly the CW input signal 221, which can have pulses with peak intensities many times higher than those present at the input. In semiconductors, the diameter D of the MRR ranges from several 100 µm to several µm. The round-trip time (Umlaufzeit) of the light here determines the repetition rate (frep) of the output signal 222.

[0055] Such microring resonators can possess a high quality factor, resulting in a peak intensity within 240° of the microring, which can drive nonlinear optical processes (so-called multiphoton processes). These can occur during the interaction of light, high intensity, and matter. Effects such as frequency harmonics or sum-difference frequency generation require two photons and induce second-order nonlinearity in the material. Third-order effects, such as third harmonics, require three photons for third-order frequency conversion, etc. These effects of nonlinear light-matter interactions provide the possibility of nonlinearly modulating incident light waves.

[0056] In the MRR, with sufficient coupling within the ring, the nonlinear refractive index becomes non-negligible. Thus, for example, due to the Kerr-Effekt effect, a four-wave mixing process occurs during the interaction of high-peak-intensity light with waveguide 210. This degenerate four-wave mixing process occurs first due to the increasing intensity within the resonator ring. Here, two photons of the input signal 221 are absorbed (optically pumped), and an electron is boosted to a higher, either virtual or real, energy level. After a short time (excitation), the electron returns to its ground state. Here, it emits the absorbed energy as signal and idler sideband photons, the sum of which is consistent with the photon energy of the two photons of the input signal 221. Therefore, a new spectral fraction is generated within the resonator.

[0057] The signal and idler sideband photons are correlated in phase, amplitude, and frequency through a coherent formation process. As the frequency transition increases, the resonator becomes bistable, resulting in slight changes in phase and frequency, which in turn generate new sidebands. A non-degenerate four-wave mixing process is used, and the generation of new frequencies is cascaded. The newly generated frequencies are in fixed phase and frequency relationships with each other, thus the spectral modes are coupled. Due to this mode coupling, fundamental solitons may develop, thereby constructing pulses with high spectral bandwidths that propagate dispersionlessly in the resonator and replicate at the resonator frequency frep. Therefore, the pulsed output signal 222, derived from the input signal 221 (e.g., a CW laser), is characterized by extremely high signal-to-noise ratio and low time variation.

[0058] Other, more complex waveguide structures can be used to generate the output signal 222 in the form of a pulse train. For example, a second waveguide 212 on the opposite side of the resonator ring can be used to decouple the pulse train (see...). Figure 5 Additional resonator rings with coupling points for further coupling between microring resonators allow tuning of the frep's frequency range. (e.g.) Figure 6 As shown, the ring component can have different diameters D, such as D1 and D2, to generate pulses with, for example, frep = 100 MHz. Other components with pulse repetition rates in the high GHz range are conceivable.

[0059] In particular, operating the MRR close to the nonlinear response threshold is advantageous for use as an optical neuron 251. Thus, the power threshold inside the resonator used to generate pulses with several µW is low, which places lower demands on the input power. Furthermore, the losses inside the resonator form an upper limit on the maximum achievable power inside the resonator, allowing the power curve to form an S-shaped function proportional to the decoupled radiation intensity. Therefore, the output signal 222 of the MRR can be used as the input signal 221 of another layer of MRR, enabling the construction of a multi-layered MRR-based optical neuron network 200.

[0060] A schematic diagram of a possible implementation of an MRR-based optical neural network 200 (MRR-ONN) is shown in Figure 2 As shown in the image, input information 231 (e.g., sensor data) is modulated onto an optical carrier via the photoelectric interface component 20 and used as input signal 221 for the first layer of MRR. Different MRRs form individual neurons 251 of the network 200, arranged in multiple layers. The output signals 222 of the MRR neurons arranged in front of them can be supplied to each individual layer via a beam splitter. The weights 252 of individual neurons 251 can be modified via optical amplifiers / attenuators. The classified output signal 222 can then be photoelectrically converted and, for example, transmitted to an environmental model.

[0061] Figure 8 It shows the relationship with Figure 7 This is similar to an MRR-ONN. However, in this case, environmental detection (i.e., input information 231) is directly guided into the MRR as input signal 221. Thus, radiation from, for example, optical devices, environmental images, or lidar systems can be directly coupled into the linear waveguide 210 of the MRR array. The spectral fraction for nonlinear operation of the MRR can be increased in intensity by means of a corresponding optical amplifier, thereby activating the nonlinear response function. This results in a neuronal pixel array of the environment without the need for photoelectric conversion of the sensor data.

[0062] The above explanation of the embodiments describes the invention only within the scope of examples. Clearly, the various features of the embodiments can be freely combined with each other without departing from the scope of the invention, as long as it is technically meaningful.

[0063] Reference Symbol List 1 System 5 Vehicle Components 6. Detection device 7 Driver Assistance Systems 8 vehicles 10 devices 20 Interface Components 200 artificial neural networks 210 Waveguide 211 Interaction Region 212 Second Wave Conductor 220 neuronal components 221 Input signal 222 Output Signal 231 Input Information 232 Output Information 240 micro resonator 241 First Micro-Resonator 242 Second Micro-resonator 243 Circular Wave Conductor 251 neurons 252 weights 101- 102 Methods and Steps A distance D diameter

Claims

1. A system (1) for a vehicle (8), comprising: - A device (10) for providing an artificial neural network (200), comprising: - At least one optical neuron component (220) for providing neurons (251) of the network (200), in, The neuron component (220) has a microresonator (240) to output the output signal (222) of the neuron component (220) in a non-linear manner related to the input signal (221) of the neuron component (220). - At least one vehicle component (5) The device (10) has an electronic and / or optoelectronic interface assembly (20) for: - Receive electrical input information (231) from the vehicle component (5), - Based on the received input information (231), the neural network (200) is provided with optical input signals (221) for the neuronal components (220). - Provide electrical output information (232) to the vehicle component (5) based on the optical output signal (222) of the neuron component (220). At least one vehicle component (5) has a detection device (6) to generate input information (231) in the form of image information, and at least one vehicle component (5) has a driver assistance system (7) for providing autonomous driving functions, so as to evaluate the output information (232) by the driver assistance system (7) and use the output information (232) as a classification of the environment of the vehicle (8). The neuron component (220) has at least one additional microresonator (240), wherein the microresonators (240) are arranged adjacent to each other, and the distance between the microresonators (240) and / or the diameter of the microresonators (240) defines a weight (252).

2. The system (1) according to claim 1, Its features are, The microresonator (240) is configured as a microring resonator.

3. The system (1) according to claim 1 or 2, Its features are, The neuron component (220) has at least one or exactly one additional microresonator (240), wherein the microresonators (240) are arranged adjacent to each other to form exactly one neuron (251).

4. The system (1) according to claim 1 or 2, Its features are, The microresonator (240) is implemented to provide an optical nonlinear effect in order to define a nonlinear relationship between the input signal (221) and the output signal (222).

5. The system (1) according to claim 1 or 2, Its features are, The neuron component (220) is configured to nonlinearly transform the input signal (221) of the neuron component (220) by means of at least one optical nonlinear effect to generate the output signal (222) of the neuron component (220), and to provide activation function with the input signal (221) of the neuron component (220) as input by means of at least one optical nonlinear effect.

6. The system (1) according to claim 1 or 2, Its features are, The neuron component (220) is configured to provide at least one weight (252) to the network (200) so as to output the output signal (222) according to the weighted sum of the input signal (221).

7. The system (1) according to claim 6, Its features are, The weights (252) are configured to linearly transform the input signal (221) to produce the output signal (222), wherein the weights (252) are defined at least in part by the arrangement and / or configuration of the microresonator (240).

8. A method for providing an artificial neural network (200), wherein the following steps are performed: - The neurons (251) of the network (200) are provided by at least one optical neuron component (220), wherein the neuron component (220) has a microresonator (240). - The output signal (222) of the neuron component (220) is output at least partially through the microresonator (240) in a manner nonlinearly related to the input signal (221) of the neuron component (220). in, The system (1) according to any one of claims 1 to 7 is provided for implementing the steps.

9. The method according to claim 8, Its features are, in, The nonlinear correlation is provided by the input signal (221) being nonlinearly converted by the microresonator to generate the output signal (222), wherein the microresonator (240) is used for this purpose by self-phase modulation.

10. The method according to claim 8 or 9, Its features are, The amplitude or frequency of the input signal (221) is non-linearly changed by the neuron (251) to obtain the output signal (222).

Images