阅读说明：本技术 一种基于机器学习的电流源注入模型建立方法 (Current source injection model establishment method based on machine learning ) 是由 李磊 袁世伟 李进 李曼 周婉婷 于 2021-07-27 设计创作，主要内容包括：本发明公开了一种基于机器学习的电流源注入模型的建立方法,具体包括：通过3D TCAD建模仿真获得机器学习所需的数据集；基于器件物理和数学变换给出注入电流注入模型的主体形式；选择时间信息t获得机器学习所需的训练集；选择f(r)机器学习模型；基于所产生的训练集进行训练和优化,得出f(r)模型所对应的参数；基于得到的f(r)参数恢复出电流源注入模型。本发明的电流源模型的建立方法基于机器学习模型和算法,通过3D TCAD仿真所得数据集,对主体模型中的核心表达式进行优化建模,从而可以应对先进工艺下的物理特性。(The invention discloses a method for establishing a current source injection model based on machine learning, which specifically comprises the following steps: obtaining a data set required by machine learning through 3D TCAD modeling simulation; giving a main body form of an injection current injection model based on device physics and mathematical transformation; selecting time information t to obtain a training set required by machine learning; selecting f (r) a machine learning model; training and optimizing based on the generated training set to obtain parameters corresponding to the f (r) model; and recovering the current source injection model based on the obtained f (r) parameters. The establishing method of the current source model is based on a machine learning model and an algorithm, and carries out optimization modeling on a core expression in a main body model through a data set obtained by 3D TCAD simulation, so that the method can cope with physical characteristics under an advanced process.)

1. A current source injection model establishing method based on machine learning comprises the following specific steps:

s1, obtaining a data set required by machine learning through 3D TCAD modeling simulation;

s2, giving a main body form of the injection current injection model based on device physics and mathematical transformation;

s3, selecting time information t to obtain a training set required by machine learning;

s4, selecting f (r) a machine learning model;

s5, training and optimizing based on the training set generated in the step S3 to obtain parameters corresponding to the f (r) model;

and S6, recovering a specific expression of I (r, t) based on the f (r) parameter obtained in the step S5, wherein I (r, t) is the established current source injection model.

2. The method for establishing a current source injection model based on machine learning of claim 1, wherein the data set of step S1 is I (r, t) data obtained by 3D TCAD modeling simulation using injection distance as input condition, where r is injection distance and t is time variable.

3. The method for establishing a current source injection model based on machine learning according to claim 2, wherein the main form of the injection current injection model of step S2 is represented as:

wherein Q is_LLET is the linear transmission energy (as input variable, depending on the irradiation environment),D_n,pis the diffusivity of a carrier, D_nDenotes the diffusivity of electrons, D_pAnd (b) represents the diffusivity of holes, T is a fixed time amount, r is an injection distance (i.e. the distance between an ion point and a collection point), I (r) is an injection current with the injection distance of r, and f (r) is a function form required to be obtained by machine learning.

4. The method according to claim 2, wherein the training set in step S3 is I (r, t) data specifying a time variable as a fixed time variable and its corresponding input condition variable.

5. The method for establishing a current source injection model based on machine learning according to claim 3, wherein the expression of the recovered I (r, t) in step S6 is as follows:

Technical Field

The invention belongs to the technical field of semiconductor technology and integrated circuits, relates to an irradiation effect simulation technology and an anti-radiation reinforcement technology in an aerospace electron or nuclear explosion environment, and particularly relates to a single event effect evaluation technology.

Background

The Single Event effect refers to that high-energy charged particles in an irradiation environment can generate energy deposition when passing through a sensitive region of an electronic device to generate a large number of electron-hole pairs, and the electron-hole pairs are respectively collected by a corresponding n region and a corresponding p region in a drifting process to generate instantaneous pulse current so as to influence the logic state of a sensitive node of the device, wherein the phenomenon of causing the level error Upset of the node of the device is called Single Event Upset effect (SEU). In the evaluation of the single event upset effect, a current source injection method is generally adopted. Therefore, how to characterize the transient pulse current by using a current source is very important for evaluating the single event upset effect sensitivity of the semiconductor device.

The current source injection model generally used is a bi-exponential model proposed by g.c. messenger in the literature "Collection of charge on junction junctions from tracks," IEEE trans.nuclear.sci., vol.ns-29, No.6, pp.2024-2031, dec.1982, as follows:

where Q is the amount of charge collected, τ_αIs the fall time constant of the junction current, τ_βRise time constant of junction current, tau_αAnd τ_βT is a time variable depending on the process parameters. The model can evaluate the single event upset threshold for evaluating devices, but the model cannot evaluate the effect of particles on surrounding devices and is not suitable for advanced processes in some cases.

CN102982216A discloses a method for establishing a current source model based on injection distance, which is based on one-dimensional injection diffusion, and assumes that all charges are collected by the same sensitive node, unlike the actual charges collected by multiple sensitive nodes.

CN111079366A discloses a method for establishing a current source model facing charge sharing, which is based on a two-dimensional diffusion idea, introduces an injection distance and a reference distance, and solves the charge sharing problem through the combined action of the injection distance and the reference distance.

Under the nano process, the physical characteristics of the semiconductor device become more complex, and even new physical characteristics appear, and the existing model is difficult to meet the requirements.

Disclosure of Invention

The invention aims to solve the problem that the existing current source injection model cannot cope with physical characteristics under an advanced process, and provides a current source injection model establishing method based on machine learning.

The technical scheme of the invention is as follows: a current source injection model establishing method based on machine learning comprises the following specific steps:

s1, obtaining a data set required by machine learning through 3D TCAD modeling simulation;

s2, giving a main body form of the injection current injection model based on device physics and mathematical transformation;

s3, selecting time information t to obtain a training set required by machine learning;

s4, selecting f (r) a machine learning model;

s5, training and optimizing based on the training set generated in the step S3 to obtain parameters corresponding to the f (r) model;

and S6, recovering a specific expression of I (r, t) based on the f (r) parameter obtained in the step S5, wherein I (r, t) is the established current source injection model.

Further, the data set in step S1 is I (r, t) data obtained by 3D TCAD modeling simulation with the implantation distance as an input condition, where r is the implantation distance (i.e. the distance between the ion point and the collection point), and t is a time variable.

Further, the main form of the injection current injection model in step S2 is represented as:

wherein Q is_LLET is linear transmission, LET 10(LET)Energy (as an input parameter, depending on the irradiation environment),D_n,pis the diffusivity of a carrier, D_nDenotes the diffusivity of electrons, D_pAnd (b) represents the diffusivity of holes, T is a fixed time amount, r is an injection distance (i.e. the distance between an ion point and a collection point), I (r) is an injection current with the injection distance of r, and f (r) is a function form required to be obtained by machine learning.

Further, the training set in step S3 is I (r, t) data with a time variable designated as a fixed time variable and corresponding input condition variables.

Further, the expression of the recovered I (r, t) in step S6:

the invention has the beneficial effects that: the establishing method of the current source model is based on a machine learning model and an algorithm, and carries out optimization modeling on a core expression in a main body model through a data set obtained by 3D TCAD simulation, so that the method can cope with physical characteristics under an advanced process.

Drawings

FIG. 1 is a flow chart of the present invention.

Fig. 2 shows a combination of a current source injection model and an NMOS circuit according to an embodiment of the present invention, wherein the arrow indicates the current direction.

Fig. 3 shows a combination of a current source injection model and a PMOS circuit according to an embodiment of the present invention, wherein the arrow indicates the current direction.

FIG. 4 shows a 6T SRAM cell and a current source injection model according to an embodiment of the invention.

Detailed Description

The invention is further described with reference to the following figures and specific embodiments.

1. The injection current injection model with time variables was modeled as:

wherein Q is_L＝10(LET)，D_n,pIs the diffusivity of a carrier, D_nDenotes the diffusivity of electrons, D_pThe diffusion rate of the holes is shown, and r is the diffusion distance, so that the related data such as semiconductor physics can be consulted.

2. Fixed time parameter T ═ T:

3. and obtaining the relevant parameters of f (r) through machine learning training.

4. And T ≠ T data information as a test set.

5. And the finally obtained I (r, t) expression is the established current source injection model.

The process of the invention is shown in figure 1, and comprises the following specific steps:

s1, obtaining a data set required by machine learning through 3D TCAD modeling simulation;

s2, providing a main body form of a current source injection model based on device physics and mathematical transformation;

s3, selecting time information t to obtain a training set required by machine learning;

s4, selecting f (r) a machine learning model;

s5, training and optimizing based on the training set generated in the step S3 to obtain parameters corresponding to the f (r) model;

and S6, restoring the specific expression of I (r, t) based on the f (r) parameter obtained in the step S5.

Based on the current source injection model described above, fig. 2 shows the current source injection model in combination with an NMOS circuit, where the arrow is the current direction, and fig. 3 shows the current source injection model in combination with a PMOS circuit, where the arrow is the current direction.

As shown in fig. 2, the current source injection model in combination with the NMOS circuit injection model comprises two parts: the NMOS transistor and the current source model, wherein D, S, B and G are respectively the drain electrode, the source electrode, the base electrode and the grid electrode of the NMOS transistor; the connection relation is as follows: the current source model is connected between the drain and the base of the NMOS transistor, and the current direction is from the drain of the transistor to the base of the transistor.

As shown in fig. 3, the current source injection model in combination with the PMOS circuit injection model comprises two parts: as shown in fig. 3, a PMOS transistor and a current source model, wherein D, S, B, G are the drain, source, base and gate of the PMOS transistor, respectively; the connection relation is as follows: the current source model is connected between the drain and the base of the PMOS transistor, and the current direction is from the base of the transistor to the drain of the transistor.

When ions attack the corresponding transistor circuit, the circuit shown in fig. 2 and 3 can be used for simulating the equivalent line injection current source, and further, the influence of the particles on the device under a certain distance can be evaluated.

The application of the model is illustrated below in the injection simulation of a single node of a specific application example 6T SRAM standard cell:

as shown in fig. 4, VDD is the power supply, GND is the ground, and the transistors M1, M2, M3, M4, M5 and M6 are connected as shown in the figure to form a standard 6T SRAM cell, where M1, M2, M5 and M6 are NMOS transistors, M3 and M4 are PMOS transistors, W is the control input signal, B and BN are the write signals, and core _ are internal level holding nodes, which can be obtained from the relevant literature, the current I (r, T) is the current source according to the embodiment of the present invention, and r represents the distance from the ion injection point to the M1 collection point.

The specific application process for the SRAM cell is as follows:

(1) the SRAM cell is designed according to the circuit structure shown in FIG. 4;

(2) based on the result of the machine learning, a current source based on I (r, t) is built in a simulation platform of the circuit, and different injection models are selected for NMOS and PMOS according to a transistor under particle attack; without loss of generality, the transistors for selecting particle attack are M1 and M1 NMOS transistors, the connection relation of injection current source models is shown in FIG. 3, and the current source is connected across the D end and the B end of the M1 tube;

(3) setting relevant parameters of I (r, t) according to the process parameters and the particle types, and setting an injection distance r;

(4) performing circuit simulation, and observing the damage degree of the current source of I (r, t) to the logic state of the circuit at the injection distance; if the logic state of the SRAM cell flips (single event upset effect) it is indicated that the particle may affect the corresponding device within the range of implant distances.

It will be appreciated by those of ordinary skill in the art that the embodiments described herein are intended to assist the reader in understanding the principles of the invention and are to be construed as being without limitation to such specifically recited embodiments and examples. Those skilled in the art can make various other specific changes and combinations based on the teachings of the present invention without departing from the spirit of the invention, and these changes and combinations are within the scope of the invention.