System and method for deriving physical quantities

Abstract

Physical quantities indicating the environment or state outside and inside a semiconductor device are acquired without installing a dedicated sensor inside the semiconductor device. [Solution] The physical quantity derivation system comprises a semiconductor device having an optical waveguide layer that transmits input light input to the input terminal to the output terminal while refracting, scattering, or reflecting it; a light source that outputs coherent light input to the input terminal; a photodetector that outputs a detection signal corresponding to the output light output from the output terminal; and a signal processing device that derives a physical quantity indicating the environment or state outside and inside the semiconductor device based on the detection signal.

Description

[Technical Field]

[0001] The disclosed technology relates to a physical quantity derivation system and a physical quantity derivation method. [Background technology]

[0002] The following techniques are known for measuring physical quantities such as temperature inside a semiconductor device. For example, Patent Document 1 describes a semiconductor device comprising a local sensor positioned in close proximity to the measurement target area of ​​a semiconductor chip and outputting the leakage current of a transistor as a sensor signal, a conversion circuit that converts the sensor signal from the local sensor into a digital count value, and a circuit block positioned between the local sensor and the conversion circuit.

[0003] Patent Document 2 describes a temperature measuring device including an optical waveguide type device. The optical waveguide type device has a first optical waveguide and a second optical waveguide formed on a substrate, and a Mach-Zehnder interferometer configured by optically coupling the first optical waveguide and the second optical waveguide at a first optical coupler and a second optical coupler, respectively. The temperature measuring device includes a light source unit that outputs light, an incident optical system that causes the light output from the light source unit to be incident on one end of either the first optical waveguide or the second optical waveguide of the optical waveguide type device, a detection unit that detects the power of the light emitted from the other end of the first optical waveguide and the second optical waveguide of the optical waveguide type device, an exit optical system that guides the light from the other end of the first optical waveguide and the second optical waveguide of the optical waveguide type device to the detection unit, and an analysis unit that determines the temperature of the optical waveguide type device based on the detection result by the detection unit.

[0004] Patent Document 3 describes a laser speckle temperature measurement method that uses multiple speckle detection systems, each of which includes a scattering plate that scatters laser light emitted from a laser oscillator, a parallelizing lens that causes the scattered laser light scattered by the scattering plate to pass through a measurement area as a parallel beam, a focusing lens that focuses the parallel laser light that has passed through the measurement area, and an image sensor that detects the speckle pattern of the laser light focused by the focusing lens. [Prior art documents] [Patent Documents]

[0005] [Patent Document 1] Japanese Patent Publication No. 2024-072439 [Patent Document 2] Japanese Patent Publication No. 2007-163886 [Patent Document 3] Japanese Patent Publication No. 2013-064641 [Overview of the Initiative] [Problems that the invention aims to solve]

[0006] In semiconductor devices that realize desired functions, such as CMOS (Metal-Oxide-Semiconductor Field-Effect Transistors), MEMS (Micro Electro Mechanical Systems), and optical integrated circuits, it is sometimes necessary to understand the external or internal environment or state of the semiconductor device in order to achieve stable operation or control.

[0007] For example, to determine the internal temperature of a semiconductor device, one might consider placing a temperature sensor inside the device. However, the measurement range of a temperature sensor is limited to a localized area. Therefore, to understand the temperature distribution of the entire semiconductor device, it would be necessary to place multiple temperature sensors at various points in the device. In this case, the placement of circuit elements such as transistors may be restricted, or the size of the semiconductor device may have to be increased.

[0008] The disclosed technology was developed in view of the above points and aims to acquire physical quantities indicating the environment or state outside and inside a semiconductor device without installing a dedicated sensor inside the semiconductor device. [Means for solving the problem]

[0009] The physical quantity derivation system relating to the disclosed technology includes a semiconductor device having an optical waveguide layer that transmits input light input to an input terminal to an output terminal while refracting, scattering, or reflecting it; a light source that outputs coherent light input to the input terminal; a photodetector that outputs a detection signal corresponding to the output light output from the output terminal; and a signal processing device that derives a physical quantity indicating the environment or state outside and inside the semiconductor device based on the detection signal.

[0010] The signal processing device may derive at least one of the following physical quantities as the ambient temperature of the semiconductor device, the internal temperature of the semiconductor device, the internal temperature distribution of the semiconductor device, the external force applied to the semiconductor device, the internal stress of the semiconductor device, and the internal stress distribution of the semiconductor device.

[0011] The signal processing device takes the detection signal as input and derives the physical quantity using an inference model learned by machine learning that outputs the physical quantity. The inference model may be trained to minimize the error between the physical quantity output by the inference model and its correct value.

[0012] The photodetector may output a one-dimensional or two-dimensional image of the output light as the detection signal. The photodetector may also output a binary signal corresponding to the amount of light of the output light as the detection signal.

[0013] The light source may output multiple lights with different wavelengths, and the photodetector may output a detection signal for each wavelength.

[0014] The optical waveguide layer may include structures that transmit light and structures that reflect light. The structures may include dummy structures that do not contribute to the original function of the semiconductor device.

[0015] The multiple structures may be made of materials with a different refractive index than the surroundings, and the size or spacing of each of the multiple structures may be approximately the same as the wavelength of the input light.

[0016] The method for deriving physical quantities related to the disclosed technology involves introducing coherent light output from a light source into the optical waveguide layer of a semiconductor device having an optical waveguide layer that transmits input light input to the input terminal to the output terminal while refracting, scattering, or reflecting it; a photodetector outputting a detection signal corresponding to the output light output from the output terminal; and a signal processing device deriving a physical quantity indicating the environment or state outside and inside the semiconductor device based on the detection signal. [Effects of the Invention]

[0017] According to the disclosed technology, it is possible to acquire physical quantities indicating the environment or state outside and inside a semiconductor device without installing a dedicated sensor inside the semiconductor device. [Brief explanation of the drawing]

[0018] [Figure 1] This figure shows an example of the configuration of a physical quantity derivation system according to an embodiment of the disclosed technology. [Figure 2] This is a cross-sectional view showing an example of the configuration of a semiconductor device according to an embodiment of the disclosed technology. [Figure 3] This is a cross-sectional view showing an example of a dummy structure according to an embodiment of the disclosed technology. [Figure 4] This figure shows an example of an optical system configuration for introducing light output from a light source according to an embodiment of the disclosed technology into an optical waveguide layer. [Figure 5] This is a cross-sectional view showing a light source embedded inside a semiconductor device. [Figure 6] This is a perspective view showing an example of the mode of optical input / output in a semiconductor device according to an embodiment of the disclosed technology. [Figure 7] This figure shows an example of input and output data for a signal processing device according to an embodiment of the disclosed technology. [Figure 8] This figure schematically illustrates the propagation of light introduced into an optical waveguide layer according to an embodiment of the disclosed technology. [Figure 9A]This diagram schematically shows the light propagation path when no warping occurs in the semiconductor device. [Figure 9B] This diagram schematically shows the light propagation path when a semiconductor device is warped. [Figure 10] This figure shows an example of a method for training an inference model to build an inference model that predicts the ambient temperature of a semiconductor device based on a detection signal. [Figure 11] This figure shows an example of a method for training an inference model to build an inference model that predicts the internal temperature distribution of a semiconductor device based on a detection signal. [Figure 12] This figure shows an example of a method for training an inference model to build an inference model that predicts the internal temperature distribution of a semiconductor device based on a detection signal. [Figure 13] This figure shows an example of a method for training an inference model to build an inference model that predicts the external force applied to a semiconductor device based on a detection signal. [Figure 14] This figure shows an example of a method for training an inference model to build an inference model that predicts the internal stress distribution of a semiconductor device based on a detection signal. [Figure 15] This figure shows an example of a method for training an inference model to construct an inference model that predicts the ambient temperature of a semiconductor device and the external force applied to the semiconductor device based on the detected signal. [Figure 16A] This image shows the brightness and darkness pattern of the output light when the temperature controller is set to 15°C. [Figure 16B] This image shows the brightness and darkness pattern of the output light when the temperature controller is set to 45°C. [Figure 17] Figures 16A and 16B show superimposed profiles of pixel values ​​along the AA' line of the light and dark patterns, respectively. [Figure 18] This figure shows two profiles obtained at the same temperature, superimposed on each other. [Figure 19A] This is a perspective view showing an example of a structure that confines light introduced into an optical waveguide layer. [Figure 19B]This is a perspective view showing an example of a structure that confines light introduced into an optical waveguide layer. [Figure 19C] This is a perspective view showing an example of a structure that confines light introduced into an optical waveguide layer. [Modes for carrying out the invention]

[0019] Hereinafter, an example of an embodiment of the disclosed technology will be described with reference to the drawings. In each drawing, identical or equivalent components and parts will be given the same reference numerals, and redundant descriptions will be omitted.

[0020] Figure 1 shows an example of the configuration of a physical quantity derivation system 1 according to an embodiment of the disclosed technology. The physical quantity derivation system 1 has the function of deriving physical quantities that represent the environment or state in at least one of the external and internal environments of a semiconductor device.

[0021] The physical quantity derivation system 1 comprises a light source 10, a semiconductor device 20, a photodetector 30, and a signal processing device 40. The semiconductor device 20 is a semiconductor chip that is the target of physical quantity derivation in this system. The semiconductor device 20 may constitute an optical integrated circuit formed using, for example, an SOI (Silicon on Insulator) substrate.

[0022] Figure 2 is a cross-sectional view showing an example of the configuration of a semiconductor device 20. Figure 2 illustrates a structure in which a support substrate 21, a BOX layer 22 (embedded oxide film layer), and a functional layer 23 are stacked. The support substrate 21 may be, for example, a silicon substrate. The BOX layer 22 is made of an insulator such as SiO2 that has light transmittance. The functional layer 23 includes structures that show transmittance and structures that show reflectance to light output from a light source 10. These structures may be elements for realizing the intrinsic function of the semiconductor device 20 (e.g., optical signal processing). Figure 2 illustrates a configuration in which a waveguide 25 and transistor 27 made of Si, a waveguide 26 made of Si3N4, a heater 28 made of metals such as Ti and W, and an electrode 29 made of metals such as Al are provided inside the optical waveguide layer 24. These structures are embedded in an insulator such as SiO2 that has light transmittance.

[0023] The optical waveguide layer 24 is formed by the BOX layer 22 and the functional layer 23. The optical waveguide layer 24 functions as a waveguide for light output from the light source 10. Light introduced into the optical waveguide layer 24 is refracted, scattered, or reflected by structures such as waveguides 25, 26, transistors 27, heaters 28, and electrodes 29. In other words, the optical waveguide layer 24 receives input light L input to the input terminal 101. in The light is transmitted to the output terminal 102 while being refracted, scattered, or reflected.

[0024] The structures provided in the functional layer 23 are not limited to those described above, and include the input light L input to the optical waveguide layer 24. in Any structure that causes refraction, scattering, or reflection may be used. For example, the structure provided on the functional layer 23 may be a dummy structure that does not contribute to the original function of the semiconductor device 20. Figure 3 is a cross-sectional view showing an example of a dummy structure. The dummy structure is provided, for example, to ensure the uniformity or flatness of the film. Figure 3 illustrates a dummy structure consisting of multiple Si pieces 111 and multiple Si3N4 pieces 112 arranged regularly and of the same shape and size. The shape, size, and arrangement of the dummy structure may be random.

[0025] The light source 10 outputs light having coherence that is input to the input end 101 of the optical waveguide layer 24 of the semiconductor device 20. The light source 10 is typically a laser light source that outputs laser light. FIG. 4 is a diagram showing an example of the configuration of an optical system for introducing the light output from the light source 10 into the optical waveguide layer 24. As shown in FIG. 4, an optical system 15 including a collimating lens 11, an optical isolator 12, a condensing lens 13, and an optical fiber 14 may be provided between the light source 10 and the semiconductor device 20. The light output from the light source 10 is introduced into the optical waveguide layer 24 via the optical system 15. Further, as shown in FIG. 5, a laser diode as the light source 10 may be embedded inside the semiconductor device 20. Also, the optical coupling between the light source 10 and the semiconductor device 20 may be performed using photonic wire bonding.

[0026] As shown in FIG. 6, the light output from the light source 10 is input to the input end 101 of the optical waveguide layer 24 and output from the output end 102 of the optical waveguide layer 24. The light propagates through the optical waveguide layer 24 while being refracted, scattered, or reflected by a structure provided inside the optical waveguide layer 24.

[0027] The photodetector 30 outputs a detection signal S out corresponding to the output light L D output from the output end 102 of the optical waveguide layer 24. The photodetector 30 includes a plurality of light receiving elements and outputs an electrical signal corresponding to the amount of light of the output light L out as the detection signal S D . The light receiving element may be, for example, a photodiode or a phototransistor. The photodetector 30 may be a line scan camera that outputs a one-dimensional image of the output light L out . Also, the photodetector 30 may be an area scan camera that outputs a two-dimensional image of the output light L out . The optical coupling between the semiconductor device 20 and the photodetector 30 can be realized using the same method as the optical coupling between the light source 10 and the semiconductor device 20.

[0028] FIG. 7 is a diagram showing an example of the input / output data of the signal processing device 40. The signal processing device 40 processes the detection signal S DBased on this, physical quantities representing the environment or state outside and inside the semiconductor device 20 are derived. Examples of physical quantities that the signal processing device 40 can derive include the ambient temperature T of the semiconductor device 20. a The internal temperature T of the semiconductor device 20 i The internal temperature distribution T of the semiconductor device 20 d External force P applied to semiconductor device 20 o , internal stress P of semiconductor device 20 i The internal stress distribution P of the semiconductor device 20 d That is the case.

[0029] The signal processing device 40 receives the detected signal S D The signal processing device 40 consists of a computer that performs the process of deriving the above physical quantities based on the detected signal S D The above physical quantities are derived using an inference model 41 trained by machine learning, which takes the above physical quantities as input and outputs them. The inference model 41 is constructed, for example, by a DNN (Deep Neural Network), and is trained to minimize the error between the physical quantities output by the model and their correct values. The training method for the inference model 41 will be described later.

[0030] The following describes the effect of heat and force generated inside and outside the semiconductor device 20 and acting on the semiconductor device 20 on the light propagating inside the optical waveguide layer 24. As shown in Figure 8, light introduced into the optical waveguide layer 24 of the semiconductor device 20 propagates towards the output terminal while repeatedly reflecting and refractioning within the optical waveguide layer 24. Due to the structure provided inside the optical waveguide layer 24, as well as the roughness and non-uniformity of the surface of the structure, the light interferes under an irregular phase relationship, resulting in the output light L out This is reflected in the light and dark patterns.

[0031] For example, when the semiconductor device 20 is heated by an external heat source, the refractive index of each element inside the optical waveguide layer 24 changes. The relationship between the change in refractive index Δn and the change in temperature ΔT of the semiconductor device 20 is expressed by equation (1) below. In equation (1), dn / dT is the refractive index temperature coefficient of each element inside the optical waveguide layer 24.

number

[0032] Furthermore, when the semiconductor device 20 is heated, the optical path length of the light propagating through the optical waveguide layer 24 changes. The temperature coefficient of the optical path length, ds / dT, is expressed by equation (2) below. In equation (2), n is the refractive index of each element inside the optical waveguide layer 24, α is the coefficient of linear expansion, and dn / dT is the refractive index temperature coefficient.

number

[0033] If the optical waveguide layer 24, which includes structures, has a temperature coefficient and a linear expansion coefficient, the refractive index and optical path length change with temperature changes in the semiconductor device 20, and the path of light propagating through the optical waveguide layer 24 changes. As a result, the output light L output from the optical waveguide layer 24 is altered. out The light and dark patterns change.

[0034] Furthermore, in the etching process, which is one of the manufacturing processes for the semiconductor device 20, fine irregularities of several nanometers are formed on the surface of the optical waveguide layer 24 (the interface between the BOX layer 22 and the functional layer 23). Light incident on this fine irregular surface is scattered at various angles. When the refractive index and optical path length of the optical waveguide layer 24 change, the intensity and phase of the scattered light change, and the output light L output from the optical waveguide layer 24 is generated. out The light and dark patterns change.

[0035] Figure 8 illustrates the case where the heat source is located outside the semiconductor device 20. However, even when the heat source is located inside the semiconductor device 20, that is, when temperature changes occur due to internal heat generation in the semiconductor device 20, the output light L output from the optical waveguide layer 24 can also be affected. out The light and dark patterns change. Internal heat generation in the semiconductor device 20 is caused by supplying power and control signals to the semiconductor device 20 to operate it. The above explanation describes the case of heating the semiconductor device 20, but the same applies to the case of cooling the semiconductor device 20.

[0036] Furthermore, when an external force is applied to the semiconductor device 20, stress is applied to each element inside the optical waveguide layer 24, causing changes in the refractive index and optical path length. Depending on the material of the structure, the stress may cause birefringence, resulting in a phase difference in polarization. Also, warping occurs in the semiconductor device 20, changing the propagation path and optical path length of the light propagating through the optical waveguide layer 24. Figure 9A shows the light propagation path when the semiconductor device 20 is not warped, and Figure 9B shows the light propagation path when the semiconductor device 20 is warped. Due to these changes associated with the application of an external force to the semiconductor device 20, the output light L output from the optical waveguide layer 24 is affected. out The light and dark pattern changes. The above explanation described the case when an external force is applied to the semiconductor device 20, but the same applies when internal stress occurs in the semiconductor device 20. That is, the refractive index, optical path length, and propagation path change due to the internal stress, and the output light L output from the optical waveguide layer 24 changes. out The light and dark patterns change. Internal stress may occur, for example, when the ambient temperature changes, due to differences in the coefficients of linear expansion between the constituent members of the semiconductor device 20.

[0037] As described above, the heat and force generated inside and outside the semiconductor device 20 and acting on the semiconductor device 20 affect the refractive index and optical path length inside the optical waveguide layer 24, and the output light L output from the optical waveguide layer 24 out This is reflected in the light and dark patterns. Output light L out By analyzing the light and dark patterns, the ambient temperature T of the semiconductor device 20 can be determined. a The internal temperature T of the semiconductor device 20 i The internal temperature distribution T of the semiconductor device 20 d External force P applied to semiconductor device 20 o , internal stress P of semiconductor device 20 i The internal stress distribution P of the semiconductor device 20 d It is possible to predict. Output light L out The light and dark pattern is detected by the photodetector 30 as an electrical signal S D It is converted and supplied to the signal processing device 40.

[0038] The training method for inference model 41 is described below. Figure 10 shows the detected signal S. D Based on the ambient temperature T of the semiconductor device 20 a This figure shows an example of a method for training an inference model 41 to build an inference model 41 that predicts [something].

[0039] Ambient temperature T of semiconductor device 20 a The temperature is set by a temperature controller 50. The temperature controller 50 includes a heater or Peltier element and heats or cools the semiconductor device 20 according to the set temperature T1. The set temperature T1 is the ambient temperature T of the semiconductor device 20. a It can be considered to be the same as.

[0040] Light emitted from the light source 10 propagates through the optical waveguide layer 24 of the semiconductor device 20. The ambient temperature of the semiconductor device 20 is T a The output light L is output from the optical waveguide layer 24. out This is reflected in the light and dark patterns. Output light L out The light and dark patterns are converted into an electrical signal by the photodetector 30, and the detected signal S D It is supplied to the signal processing device 40 as the output light L. out An area scan camera that outputs a two-dimensional image can be used.

[0041] The signal processing device 40 receives the detected signal S D The detected signal S is input to the inference model 41. D Based on the ambient temperature T a The predicted value T2 is output. The inference model 41 is composed of a DNN with updatable parameters. The parameters inside the DNN are updated so that the error E between the set temperature T1 (ground truth value) of the temperature controller 50 and the predicted value T2 output from the inference model 41 is minimized. This enables the learning of the inference model 41.

[0042] For example, output light L out Detection signal S shows a two-dimensional image of the light and dark pattern. DA dataset consisting of combinations of the set temperature T1 (ground truth value) and the corresponding set temperature T1 is used as training data for the inference model 41. Multiple datasets obtained under different set temperatures T1 are used as training data for the inference model 41.

[0043] Figure 11 shows the detected signal S D Based on the internal temperature distribution T of the semiconductor device 20 d This figure shows an example of a method for training the inference model 41 to construct an inference model 41 that predicts the following. The method shown below is an example where the heat source is outside the semiconductor device 20.

[0044] The internal temperature distribution of the semiconductor device 20 is formed by the temperature controller 50. The temperature controller 50 includes a heater or a Peltier element and can form a desired temperature distribution inside the semiconductor device 20 by heating or cooling the semiconductor device 20 at different temperatures in different regions.

[0045] Internal temperature distribution T of semiconductor device 20 d This is visualized by a far-infrared camera 60 installed above the semiconductor device 20. In other words, the far-infrared camera 60 visualizes the internal temperature distribution T of the semiconductor device 20. d Temperature distribution image I showing td1 Outputs: Temperature distribution image I td1 This is an image showing the internal temperature distribution of the semiconductor device 20, displayed in different colors.

[0046] Light emitted from the light source 10 propagates through the optical waveguide layer 24 of the semiconductor device 20. The internal temperature distribution T of the semiconductor device 20. d The output light L is output from the optical waveguide layer 24. out This is reflected in the light and dark patterns. Output light L out The light and dark patterns are converted into an electrical signal by the photodetector 30, and the detected signal S D It is supplied to the signal processing device 40 as the output light L. out An area scan camera that outputs a two-dimensional image can be used.

[0047] The signal processing device 40 receives the detected signal S D The detected signal S is input to the inference model 41. D Based on this, the internal temperature distribution T of the semiconductor device 20 d Predicted temperature distribution image I showing the prediction results td2 The output is a temperature distribution image I output from the far-infrared camera 60. The inference model 41 is composed of a DNN with updatable parameters. td1 (Correct value) and the predicted temperature distribution image I output from inference model 41 td2 The parameters within the DNN are updated to minimize the error E. This enables the training of the inference model 41.

[0048] For example, output light L out Detection signal S shows a two-dimensional image of the light and dark pattern. D And the corresponding temperature distribution image I td1 A dataset consisting of combinations with (correct values) is used as training data for the inference model 41. The internal temperature distributions T are different from each other. d Multiple datasets obtained under the specified conditions are used as training data for the inference model 41.

[0049] According to the inference model 41 trained by the above method, the internal temperature distribution T of the semiconductor device 20 d Furthermore, the internal temperature T in a specific part of the semiconductor device 20 i It is also possible to predict this.

[0050] Figure 12 shows the detected signal S. D Based on the internal temperature distribution T of the semiconductor device 20 d This figure shows an example of a method for training an inference model 41 to build an inference model 41 that predicts the following. The method shown below is an example where the heat source is inside the semiconductor device 20.

[0051] The internal temperature distribution of the semiconductor device 20 is controlled by the control signal S of the semiconductor device 20. CThis is formed by supplying heat and operating the semiconductor device 20. Other aspects are the same as when the heat source is located outside the semiconductor device 20 (see Figure 11), so further explanation is omitted.

[0052] Figure 13 shows the detected signal S. D An external force P applied to the semiconductor device 20 based on O This figure shows an example of a method for training an inference model 41 to build an inference model 41 that predicts [something].

[0053] External force is applied to the semiconductor device 20 by the pressure application device 70. The pressure application device 70 applies a pressing force to the semiconductor device 20 according to the set pressure P1. The set pressure P1 is the external force P applied to the semiconductor device 20. O It can be considered to be the same as.

[0054] Light emitted from the light source 10 propagates through the optical waveguide layer 24 of the semiconductor device 20. An external force P is applied to the semiconductor device 20. O The output light L is output from the optical waveguide layer 24. out This is reflected in the light and dark patterns. Output light L out The light and dark patterns are converted into an electrical signal by the photodetector 30, and the detected signal S D It is supplied to the signal processing device 40 as the output light L. out An area scan camera that outputs a two-dimensional image can be used.

[0055] The signal processing device 40 receives the detected signal S D The detected signal S is input to the inference model 41. D Based on this, the predicted value P2 of the external force applied to the semiconductor device 20 is output. The inference model 41 is composed of a DNN with updatable parameters. The parameters inside the DNN are updated so that the error E between the set pressure P1 (ground truth value) of the pressure application device 70 and the predicted value P2 output from the inference model 41 is minimized. This enables the learning of the inference model 41.

[0056] For example, output light Lout The detection signal S showing the two-dimensional image of the light and dark pattern D A data set consisting of a combination with the corresponding set pressure P1 (correct value) is used as learning data for the inference model 41. A plurality of data sets obtained under different set pressures P1 are used as learning data for the inference model 41.

[0057] FIG. 14 D Based on the internal stress distribution P of the semiconductor device 20 d is a diagram showing an example of a learning method of the inference model 41 for constructing the inference model 41 for predicting.

[0058] The internal stress distribution P of the semiconductor device 20 d is formed by the temperature controller 50. The temperature controller 50 includes a heater or a Peltier element and heats or cools the semiconductor device 20 according to the set temperature T1. By heating or cooling the semiconductor device 20, a stress distribution is formed inside the semiconductor device 20 due to the difference in the linear expansion coefficient between the constituent members of the semiconductor device 20.

[0059] The internal stress distribution P of the semiconductor device 20 d is visualized by the stress simulator 80. Information such as the set temperature T1 of the temperature controller 50, the internal structure model (CAD data) of the semiconductor device 20, the dimensions, Young's modulus, Poisson's ratio, density, and linear expansion coefficient of each component of the semiconductor device 20 is input to the stress simulator 80. Based on these input information, the stress simulator 80 outputs a stress distribution image I d showing the internal stress distribution P of the semiconductor device 20 pd1 The stress distribution image I pd1 is an image in which the magnitude of stress is displayed in color-coded form.

[0060] The light output from the light source 10 propagates through the optical waveguide layer 24 of the semiconductor device 20. The internal stress distribution P of the semiconductor device 20 d is reflected in the light and dark pattern of the output light L out output from the optical waveguide layer 24. The output light L outThe light and dark pattern is converted into an electrical signal by the photodetector 30 and supplied to the signal processing device 40 as the detection signal S. D As the photodetector 30, an area scan camera that outputs a two-dimensional image of the output light L out can be used.

[0061] The signal processing device 40 inputs the detection signal S D to the inference model 41. Based on the detection signal S D , the inference model 41 outputs a predicted stress distribution image I d indicating the prediction result of the internal stress distribution P pd2 of the semiconductor device 20. The inference model 41 is composed of a DNN having updatable parameters. The stress distribution image I pd1 (correct value) output from the stress simulator 80 and the predicted stress distribution image I pd2 output from the inference model 41, the parameters inside the DNN are updated so that the error E is minimized. Thereby, the learning of the inference model 41 is realized.

[0062] For example, a dataset consisting of a combination of the detection signal S out showing a two-dimensional image of the light and dark pattern of the output light L and the corresponding stress distribution image I D (correct value) is used as the learning data of the inference model 41. A plurality of datasets obtained under different internal stress distributions P pd1 are used as the learning data of the inference model 41. d According to the inference model 41 learned by the above method, not only the internal stress distribution P d of the semiconductor device 20 but also the internal stress P i at a specific part of the semiconductor device 20 can be predicted. d Figure 15 shows the ambient temperature T i of the semiconductor device 20 and the external force P

[0064] based on the detection signal S D applied to the semiconductor device 20. a and the external force P OThis figure shows an example of a method for training an inference model 41 to construct an inference model 41 that predicts each of the following.

[0065] Ambient temperature T of semiconductor device 20 a The temperature is set by a temperature controller 50. The temperature controller 50 includes a heater or Peltier element and heats or cools the semiconductor device 20 according to the set temperature T1. The set temperature T1 is the ambient temperature T of the semiconductor device 20. a It can be considered to be the same as.

[0066] External force is applied to the semiconductor device 20 by the pressure application device 70. The pressure application device 70 applies a pressing force to the semiconductor device 20 according to the set pressure P1. The set pressure P1 is the external force P applied to the semiconductor device 20. O It can be considered to be the same as.

[0067] Light emitted from the light source 10 propagates through the optical waveguide layer 24 of the semiconductor device 20. The ambient temperature of the semiconductor device 20 is T a and the external force P applied to the semiconductor device 20 O The output light L is output from the optical waveguide layer 24. out This is reflected in the light and dark patterns. Output light L out The light and dark patterns are converted into an electrical signal by the photodetector 30, and the detected signal S D It is supplied to the signal processing device 40 as the output light L. out An area scan camera that outputs a two-dimensional image can be used.

[0068] The signal processing device 40 receives the detected signal S D The detected signal S is input to the inference model 41. D Based on the ambient temperature T a The predicted value T2 and the external force P applied to the semiconductor device 20 OThe predicted value P2 is output. The inference model 41 is composed of a DNN with updatable parameters. The parameters inside the DNN are updated so that the error E1 between the set temperature T1 (ground truth value) of the temperature controller 50 and the predicted value T2 output from the inference model 41 is minimized. In addition, the parameters inside the DNN are updated so that the error E2 between the set pressure P1 (ground truth value) of the pressure application device 70 and the predicted value P2 output from the inference model 41 is minimized. This enables the learning of the inference model 41.

[0069] For example, output light L out Detection signal S shows a two-dimensional image of the light and dark pattern. D Then, a dataset consisting of combinations of the corresponding set temperature T1 (ground truth value) and the corresponding set pressure P1 (ground truth value) is used as training data for the inference model 41. Multiple datasets obtained under different combinations of set temperature T1 and set pressure P1 are used as training data for the inference model 41.

[0070] The signal processing device 40 operates at an ambient temperature T a , internal temperature T i , internal temperature distribution T d , external force P o Internal stress P i and internal stress distribution P d It may have multiple inference models 41 that individually derive the results.

[0071] The operation of the physical quantity derivation system 1 is described below. The light source 10 emits coherent light. The light emitted from the light source 10 is introduced into the optical waveguide layer 24 of the semiconductor device 20. Input light L to the optical waveguide layer 24 in The light propagates through the optical waveguide layer 24, repeatedly undergoing refraction, scattering, or reflection. The heat and force generated inside and outside the semiconductor device 20 and acting on the semiconductor device 20 affect the refractive index and optical path length within the optical waveguide layer 24, and the output light L is output from the optical waveguide layer 24. out This is reflected in the light and dark patterns. Output light L out The light and dark patterns are converted into an electrical signal by the photodetector 30, and the detected signal S DThe signal is supplied to the signal processing device 40 as detected signal S. D The detected signal S is input to the inference model 41. D Based on this, physical quantities representing the environment or state outside and inside the semiconductor device 20 are derived. The signal processing device 40 determines the ambient temperature T a , internal temperature T i , internal temperature distribution T d , external force P o Internal stress P i and internal stress distribution P d It is possible to derive at least one of these as a physical quantity.

[0072] [Examples] Using the system with the configuration illustrated in Figure 10, the output light L output from the optical waveguide layer 24 is used. out The following was observed. Laser light with a wavelength of 543.5 nm output from light source 10 was introduced into the optical waveguide layer 24 of semiconductor device 20. Semiconductor device 20 is an optical integrated circuit manufactured using an SOI substrate. The optical waveguide layer 24 receives the input light L in Waveguides made of Si and Si3N4 and a metal film made of Al are provided as structures that cause refraction, scattering, or reflection of light. Output light L from the optical waveguide layer 24 out The light and dark patterns were captured using a CMOS (Complementary Metal Oxide Semiconductor) image sensor.

[0073] Figures 16A and 16B show the output light L, respectively. out These are images of the light and dark patterns. Figure 16A shows the image when the temperature controller 50 is set to 15°C, and Figure 16B shows the image when the temperature controller 50 is set to 45°C. Figure 17 is a diagram showing the pixel value profiles along the AA' line of the light and dark patterns shown in Figures 16A and 16B, respectively, superimposed on each other. The two profiles are different from each other. This indicates that the ambient temperature is different from the output light L from the optical waveguide layer 24. out This suggests that it will be reflected in the light and dark patterns.

[0074] Next, an image of the brightness and darkness pattern of the output light was captured when the temperature controller 50 was set to 25°C, and the same profile as above was obtained for that image (1st time). Furthermore, after raising the temperature controller 50 to 45°C, the temperature was returned to 25°C and the output light L was captured after the temperature stabilized. out Images of the light-dark pattern were captured, and the same profile as above was obtained for those images (second time). Figure 18 shows the first and second profiles superimposed. The two profiles are almost identical. This suggests that the reflection of ambient temperature on the light-dark pattern is reproducible.

[0075] As described above, the physical quantity derivation system 1 according to the embodiment of the disclosed technology receives input light L input to the input terminal 101. in A semiconductor device 20 having an optical waveguide layer 24 that transmits light to the output terminal 102 while refracting, scattering, or reflecting it, a light source 10 that outputs coherent light input to the input terminal 101, and output light L output from the output terminal 102 out Detection signal S corresponding to the detection signal S D A photodetector 30 that outputs a detection signal S D The semiconductor device 20 includes a signal processing device 40 that derives a physical quantity indicating the environment or state outside and inside the semiconductor device 20 based on this.

[0076] According to the semiconductor device 20 of the disclosed technology, it is possible to acquire physical quantities indicating the environment or state outside and inside the semiconductor device 20 without providing a dedicated sensor inside the semiconductor device 20. Since it is not necessary to provide a dedicated sensor inside the semiconductor device 20, the arrangement of circuit elements such as transistors provided inside the semiconductor device 20 is not restricted, and there is no need to increase the size of the semiconductor device 20.

[0077] Furthermore, according to the semiconductor device 20 of this embodiment, it is possible to derive physical quantities not only in a localized area but also over a wide area. Specifically, it is possible to derive the internal temperature distribution and internal stress distribution over the area affected by light propagating through the optical waveguide layer 24.

[0078] [Differentiation] Various modifications can be made to the physical quantity derivation system 1 described above. For example, the light source 10 may output multiple lights with different wavelengths in a switchable manner. To achieve this, the light source 10 may be equipped with multiple laser diodes. In addition, the photodetector 30 may have a spectral function, such as a visible light (RGB) camera, and detect a signal S for each wavelength. D It may also output the detection signal S. D Since color information can be included, the amount of input information to the signal processing device 40 can be increased, making it possible to improve the accuracy of derivation of physical quantities.

[0079] Furthermore, the photodetector 30 emits output light L out A binary signal corresponding to the light intensity is detected as the detection signal S. D The system may also have a SPAD (Single Photon Avalanche Diode) as a photodetector that outputs as S. D By using a binary signal, the amount of input information to the signal processing device 40 can be reduced, thereby reducing the processing cost (e.g., processing time, power consumption) of the signal processing device 40.

[0080] Furthermore, while DNN was used as an example of inference model 41 in the above explanation, the disclosed technologies are not limited to this. Inference model 41 may be constructed using any machine learning approach.

[0081] Furthermore, while the above description illustrates a configuration in which the signal processing device 40 derives physical quantities using an inference model 41 learned by machine learning, the disclosed technology is not limited to this. DThe physical quantity may be derived by performing pattern matching. Specifically, the detected signal S D Multiple datasets, each consisting of a pair of a physical quantity and a corresponding physical quantity, are pre-stored in internal memory. The newly acquired detection signal S D Regarding the detection signal S stored in the internal memory D The physical quantity corresponding to the most similar of these is the newly acquired detection signal S. D It is derived as a physical quantity corresponding to the signal S. Furthermore, the signal processing device 40 processes the detected signal S. D An alert may be issued if the difference between the light / dark pattern shown and a pre-defined standard light / dark pattern exceeds a threshold.

[0082] Furthermore, the optical waveguide layer 24 may have a structure that confines the introduced light. Figure 19 is a perspective view showing an example of a structure that confines the light introduced into the optical waveguide layer 24. Multiple structures 120 are formed inside the optical waveguide layer 24 to confine the input light L in They are arranged at both ends of the path along the path. Structure 120 is made of a material with a different refractive index from the surroundings. For example, if the material (medium) surrounding structure 120 is Si, SiO2 can be used as the material for structure 120. The size or spacing of each of the multiple structures 120 is determined by the input light L in The wavelength is approximately the same as (for example, input light L in The refractive index is said to be approximately plus or minus 50% of the wavelength. Multiple structures 120 form a periodic structure in which materials with different refractive indices are arranged at intervals similar to the wavelength of light. As a result, the multiple structures 120 function as photonic crystals, resulting in the effect of confining the light introduced into the optical waveguide layer 24. In other words, the light introduced into the optical waveguide layer 24 cannot penetrate outside the region sandwiched between the structures 120.

[0083] By providing a structure that confines light in the optical waveguide layer 24, it becomes possible to promote the interaction between the light introduced into the optical waveguide layer 24 and the internal structures of the optical waveguide layer 24 (transistors, waveguides, electrodes, etc.). Furthermore, the output light L outThe intensity can also be improved. This makes it possible to improve the accuracy of the derivation of physical quantities in the signal processing device 40. As shown in Figures 19B and 19C, at least a portion of the multiple structures 120 are input light L in They may be arranged in a direction intersecting the direction of propagation. In this case, the output light L out The input light L in The light is folded back in the opposite direction to its direction of propagation. Even when the light introduced into the optical waveguide layer 24 is folded back within the optical waveguide layer 24, it is still possible to promote the interaction between the light introduced into the optical waveguide layer 24 and the internal structures of the optical waveguide layer 24 (transistors, waveguides, electrodes, etc.).

[0084] The following additional information is disclosed regarding the embodiments described above. (Note 1) A semiconductor device having an optical waveguide layer that transmits input light input to the input terminal to the output terminal while refracting, scattering, or reflecting it, A light source that outputs coherent light input to the aforementioned input terminal, A photodetector that outputs a detection signal corresponding to the output light output from the output terminal, A signal processing device that derives a physical quantity indicating the environment or state outside or inside the semiconductor device based on the detection signal, A system for deriving physical quantities that has [a certain characteristic].

[0085] (Note 2) The signal processing device derives at least one of the following physical quantities: the ambient temperature of the semiconductor device, the internal temperature of the semiconductor device, the internal temperature distribution of the semiconductor device, the external force applied to the semiconductor device, the internal stress of the semiconductor device, and the internal stress distribution of the semiconductor device. The physical quantity derivation system described in Appendix 1.

[0086] (Note 3) The signal processing device takes the detection signal as input and derives the physical quantity using an inference model learned by machine learning that outputs the physical quantity. The inference model is trained to minimize the error between the physical quantity output by the inference model and its correct value. A physical quantity derivation system as described in Appendix 1 or Appendix 2.

[0087] (Note 4) The photodetector outputs a one-dimensional or two-dimensional image of the output light as the detection signal. A physical quantity derivation system described in any one of the appendices 1 through 3.

[0088] (Note 5) The photodetector outputs a binary signal corresponding to the amount of light from the output light as the detection signal. A physical quantity derivation system described in any one of the appendices 1 through 3.

[0089] (Note 6) The aforementioned light source outputs multiple lights with different wavelengths, The aforementioned photodetector outputs a detection signal for each wavelength. A physical quantity derivation system described in any one of the appendices 1 through 5.

[0090] (Note 7) The optical waveguide layer includes structures that transmit light and structures that reflect light. A physical quantity derivation system described in any one of the appendices 1 through 6.

[0091] (Note 8) The aforementioned structure includes a dummy structure that does not contribute to the original function of the semiconductor device. The physical quantity derivation system described in Appendix 7.

[0092] (Note 9) Multiple of the aforementioned structures are made of materials with a different refractive index than the surroundings. The size or spacing of each of the multiple structures is approximately the same as the wavelength of the input light. A physical quantity derivation system as described in Appendix 7 or Appendix 8.

[0093] (Note 10) A semiconductor device having an optical waveguide layer that transmits input light input to the input terminal to the output terminal while refracting, scattering, or reflecting it, introduces coherent light output from a light source into the optical waveguide layer. The photodetector outputs a detection signal corresponding to the output light output from the output terminal. The signal processing device derives a physical quantity that indicates the environment or state outside or inside the semiconductor device, based on the detection signal. Physical quantity derivation method. [Explanation of Symbols]

[0094] 1. System for Deriving Physical Quantities 10 light source 20 Semiconductor equipment 24 Optical waveguide layer 30 Photodetectors 40 Signal Processing Device 41 Inference Models 120 Structures

Claims

1. A semiconductor device having an optical waveguide layer that transmits input light input to the input terminal to the output terminal while refracting, scattering, or reflecting it, A light source that outputs coherent light input to the aforementioned input terminal, A photodetector that outputs a detection signal corresponding to the output light output from the output terminal, A signal processing device that derives a physical quantity indicating the environment or state outside or inside the semiconductor device based on the detection signal, A system for deriving physical quantities that has [a certain characteristic].

2. The signal processing device derives at least one of the following physical quantities: the ambient temperature of the semiconductor device, the internal temperature of the semiconductor device, the internal temperature distribution of the semiconductor device, the external force applied to the semiconductor device, the internal stress of the semiconductor device, and the internal stress distribution of the semiconductor device. The physical quantity derivation system according to claim 1.

3. The signal processing device takes the detection signal as input and derives the physical quantity using an inference model learned by machine learning that outputs the physical quantity. The inference model is trained to minimize the error between the physical quantity output by the inference model and its correct value. The physical quantity derivation system according to claim 1.

4. The photodetector outputs a one-dimensional or two-dimensional image of the output light as the detection signal. The physical quantity derivation system according to claim 1.

5. The photodetector outputs a binary signal corresponding to the amount of light from the output light as the detection signal. The physical quantity derivation system according to claim 1.

6. The aforementioned light source outputs multiple lights with different wavelengths, The aforementioned photodetector outputs a detection signal for each wavelength. The physical quantity derivation system according to claim 1.

7. The optical waveguide layer includes structures that transmit light and structures that reflect light. The physical quantity derivation system according to claim 1.

8. The aforementioned structure includes a dummy structure that does not contribute to the original function of the semiconductor device. The physical quantity derivation system according to claim 7.

9. Multiple of the aforementioned structures are made of materials with a different refractive index than the surroundings. The size or spacing of each of the multiple structures is approximately the same as the wavelength of the input light. The physical quantity derivation system according to claim 7.

10. A semiconductor device having an optical waveguide layer that transmits input light input to the input terminal to the output terminal while refracting, scattering, or reflecting it, introduces coherent light output from a light source into the optical waveguide layer. The photodetector outputs a detection signal corresponding to the output light output from the output terminal. The signal processing device derives a physical quantity that indicates the environment or state outside or inside the semiconductor device, based on the detection signal. Physical quantity derivation method.

Images