Semiconductor failure analysis device and failure analysis method thereof

Through deep neural network and dual sampling technology, semiconductor fault analysis is combined with simulation data, the problem of semiconductor fault identification is solved, the accuracy and reliability of the analysis is improved, and process changes are adapted.

CN110619922BActive Publication Date: 2024-11-12SAMSUNG ELECTRONICS CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN201910444511.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2018-06-18
Filing Date
2019-05-27
Publication Date
2024-11-12
Estimated Expiration
2039-05-27

AI Technical Summary

Technical Problem

In the semiconductor manufacturing process, as process complexity increases, it becomes more difficult to identify and classify semiconductor failures, and the prior art is difficult to effectively improve the accuracy and reliability of fault analysis, especially in the case of insufficient training data or changes in manufacturing processes. Down.

Method used

The semiconductor fault analysis device based on deep neural network is adopted to generate dual sampling technology of measurement data and simulation data, and pre-trained with the simulation data to generate simulation data to ensure accurate classification and analysis of fault types.

Benefits of technology

Improves the accuracy and reliability of semiconductor fault analysis, reduces analysis time, and maintains high reliability when training data is insufficient or process changes, ensuring accurate classification and cause identification of fault types.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN110619922B_ABST
    Figure CN110619922B_ABST
Patent Text Reader

Abstract

A semiconductor failure analysis device and a failure analysis method thereof are provided. The failure analysis method comprises: receiving measured data corresponding to a semiconductor device; generating double sampling data based on the measured data and reference data; performing a failure analysis operation on the double sampling data; classifying a failure type of the semiconductor device based on a result of the failure analysis operation; and outputting information about the failure type.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the priority of Korean Patent Application No. 10-2018-0069584 filed on June 18, 2018 in the Korean Intellectual Property Office, the disclosure of which is incorporated herein by reference in its entirety. Technical Field

[0003] Methods and apparatuses consistent with example embodiments relate to semiconductor devices, and more particularly, to semiconductor failure analysis apparatuses and failure analysis methods thereof. Background Art

[0004] Semiconductor devices are manufactured through various processes. With the development of semiconductor design technology, the number of processes for manufacturing semiconductor devices, the complexity of each process, and the integration level are increasing. Therefore, various defects or failures may occur in the semiconductor manufacturing process.

[0005] Semiconductor failures should be identified as quickly as possible so that the cause of the failure can be identified and corrected.Semiconductor failures can be classified based on electrical measurements of semiconductor devices (eg, current-voltage characteristics). Summary of the invention

[0006] Example embodiments provide a semiconductor failure analysis apparatus having improved reliability and a failure analysis method thereof.

[0007] According to an aspect of an exemplary embodiment, a failure analysis method of a semiconductor failure analysis apparatus is provided, the failure analysis method comprising: receiving measurement data measured from a semiconductor device included in a semiconductor wafer; generating double sampling data based on the measurement data and reference data; performing a failure analysis operation on the double sampling data; classifying a failure type of the semiconductor device based on a result of the failure analysis operation; and outputting information about the failure type.

[0008] According to an aspect of an exemplary embodiment, there is provided a semiconductor failure analysis apparatus, comprising: a measuring apparatus configured to measure electrical data from a semiconductor device included in a semiconductor wafer and output the measured electrical data as measurement data; a first double sampler configured to generate double sampling data based on the measurement data and reference data; a classifier configured to perform a failure analysis operation based on the double sampling data and classify a failure type of the semiconductor device based on a result of the failure analysis operation; and an analyzer configured to perform analysis based on the failure type and output a result of the analysis.

[0009] According to an aspect of an exemplary embodiment, there is provided a semiconductor failure analysis apparatus, comprising: a measuring apparatus configured to receive electrical data from a semiconductor device included in a semiconductor wafer and output the electrical data as measurement data; a simulator configured to generate a simulation for each of a plurality of failure types of the semiconductor device and to generate simulation data based on the simulation; a classifier configured to generate a training model by performing pre-training based on the simulation data, perform a failure analysis operation on the measurement data by using the training model, and classify the failure type of the semiconductor device based on the failure analysis operation; and an analyzer configured to perform analysis based on the failure type and output a result of the analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The above and other objects and features will become more apparent from the following description in conjunction with the accompanying drawings.

[0011] Figure 1 is a block diagram illustrating a semiconductor failure analysis system according to an example embodiment.

[0012] Figure 2 is a circuit diagram illustrating a memory cell for describing an example embodiment.

[0013] Figure 3 It is an example Figure 1 Flowchart of the operation of the emulator.

[0014] FIG. 4A to FIG. 4D is a schematic diagram for describing some of the various types of failures that can occur in a memory cell of an SRAM device.

[0015] FIG. 5A to FIG. 5D is an example of FIG. 4A to FIG. 4D Schematic diagram of simulation data generated for each fault type from first to fourth fault types.

[0016] Figure 6 is a block diagram illustrating a semiconductor failure analysis system according to an example embodiment.

[0017] Figure 7 It is an example Figure 6 A flow chart of an operating method of a semiconductor failure analysis device.

[0018] FIG. 8A to FIG. 8C Is used to describe Figure 6 Schematic diagram of the operation of the semiconductor failure analysis device.

[0019] 9A to 9D Is used to describe Figure 6Schematic diagram of the operation of the semiconductor failure analysis device.

[0020] Fig.10 is a block diagram illustrating a semiconductor failure analysis system according to an example embodiment.

[0021] Fig.11 It is an example Fig.10 A flow chart of the operation of a semiconductor failure analysis apparatus.

[0022] Fig.12 is a block diagram illustrating a semiconductor failure analysis system according to an example embodiment.

[0023] Fig.13 is a schematic diagram for describing measurement data according to another example embodiment.

[0024] Fig.14 is a schematic diagram illustrating any one of various semiconductor devices fabricated on a semiconductor wafer according to an example embodiment.

[0025] Fig.15 is a schematic diagram illustrating any one of various semiconductor devices fabricated on a semiconductor wafer according to an example embodiment. DETAILED DESCRIPTION

[0026] Below, the example embodiments will be described in detail with reference to the accompanying drawings. In the following description, known configurations may be omitted. In addition, the following example embodiments may be modified into various other forms, and the scope of the technical spirit of the present disclosure is not limited to the following examples. On the contrary, these example embodiments are provided to make the present disclosure more thorough and complete, and to fully convey the scope of the technical spirit of the present disclosure to those skilled in the art.

[0027] Figure 1 is a block diagram illustrating a semiconductor failure analysis system according to an example embodiment. Figure 1 The semiconductor failure analysis system 10 may include a semiconductor wafer WF, a manufacturing device 11 , a measuring device 12 , and a semiconductor failure analysis device 100 .

[0028] The semiconductor wafer WF may be used as a substrate for a semiconductor device. For example, the semiconductor wafer WF may include a material having semiconductor properties, such as silicon (Si) or gallium arsenide (GaAs). The manufacturing device 11 may perform various processes for manufacturing the semiconductor wafer WF. For example, the manufacturing device 11 may form various semiconductor patterns on the semiconductor wafer WF by performing various processes such as an etching process, a deposition process, and a planarization process on the semiconductor wafer WF.

[0029] In an exemplary embodiment, the semiconductor wafer WF having the semiconductor pattern formed by the manufacturing device 11 may include a logic circuit that performs various logic operations, or a memory device such as a static random access memory (SRAM), a dynamic random access memory (DRAM), a flash memory, or a magnetic random access memory (MRAM). Below, for convenience, it is described that the semiconductor wafer WF includes a pattern for an SRAM device. That is, the semiconductor wafer WF may include a plurality of SRAM devices.

[0030] The measuring device 12 can measure various information from the SRAM device formed on the semiconductor wafer WF. For example, the measuring device 12 can measure current-voltage data I-V_m (hereinafter referred to as "measurement data") from the SRAM device formed on the semiconductor wafer WF. The measurement data I-V_m can be information indicating the current-voltage characteristics of each memory cell in a plurality of memory cells included in the SRAM device or information indicating the current-voltage characteristics of each semiconductor element included in each memory cell. The measurement data I-V_m will be described more fully below.

[0031] The semiconductor failure analysis apparatus 100 may include a simulator 110, a failure type database 120, a classifier 130, and an analyzer 140. In an example embodiment, the semiconductor failure analysis apparatus 100 may perform failure analysis on a semiconductor wafer WF based on an artificial intelligence (AI) algorithm such as a convolutional neural network (CNN), a deep neural network, or machine learning. Below, for ease of description, it is assumed that the semiconductor failure analysis apparatus 100 performs an analysis operation based on a deep neural network algorithm.

[0032] The classifier 130 of the semiconductor failure analysis device 100 may perform a failure analysis operation based on a deep neural network algorithm. For example, the classifier 130 may receive measurement data I-V_m from the measurement device 12. By performing a failure analysis operation on the received measurement data I-V_m, the classifier 130 may determine whether the SRAM device included in the semiconductor wafer WF is normal or faulty, and may determine the type of failure. Below, for ease of description, the operation of determining whether the SRAM device included in the semiconductor wafer WF is normal or faulty and the operation of determining the type of failure are referred to as "failure analysis operation".

[0033] For example, by extracting characteristics of the measurement data I-V_m and performing a deep neural network algorithm on the extracted characteristics, the classifier 130 of the semiconductor failure analysis apparatus 100 may classify whether the measurement data I-V_m corresponds to a normal state or a failure state, and may determine the failure type. In an example embodiment, the characteristics extracted by the classifier 130 may include the amplitude, slope, inflection point, and waveform shape of a signal associated with each of a plurality of regions of the measurement data I-V_m, but example embodiments are not limited thereto.

[0034] In an example embodiment, the classifier 130 may generate a training model by pre-training so as to perform an analysis operation based on a deep neural network algorithm. Pre-training refers to a process of simulating a fault type associated with an SRAM device of a semiconductor wafer WF based on data associated with various fault types (e.g., current-voltage characteristics). The classifier 130 may perform an analysis operation on the measurement data I-V_m based on the training model generated by pre-training.

[0035] In the case where data for pre-training (e.g., current-voltage data) or information associated with various fault types is insufficient, an accurate training model may not be generated, and thus, the operational reliability of the classifier 130 may become poor. That is, sufficient data for pre-training may be required to make the operation of the classifier 130 reliable.

[0036] Typically, data for pre-training can be obtained by performing various semiconductor manufacturing processes and performing failure analysis operations on various semiconductor wafers. However, due to various factors such as changes in manufacturing processes, changes in characteristics of semiconductor wafers, and changes in specifications of semiconductor devices, pre-training data may be insufficient.

[0037] In an example embodiment, the semiconductor failure analysis device 100 can ensure the reliability of the failure analysis operation even in the case of insufficient data for pre-training. For example, the simulator 110 of the semiconductor failure analysis device 100 can generate a simulation associated with each of the various failure types that may occur in the SRAM device included in the semiconductor wafer WF, and can generate or output current-voltage data I-V_s (hereinafter referred to as "simulation data") based on the generated simulation. In an example embodiment, the simulator 110 can be implemented in hardware or software, wherein the hardware or software is operated based on a public open source analog circuit simulation such as SPICE (simulation program within integrated circuit emphasis), or is implemented in a combination of hardware and software, but the example embodiment is not limited thereto.

[0038] The simulation data I-V_s may be stored to the fault type database 120. The fault type database 120 may be a storage medium storing various simulation data I-V_s generated by the simulator 110. The fault type database 120 may output the stored simulation data I-V_s as training data I-V_t.

[0039] The classifier 130 may generate a training model by performing pre-training based on the training data I-V_t from the fault type database 120. For example, as described above, the fault type database 120 may include simulation data I-V_s of various fault types provided from the simulator 110, and the simulation data I-V_s may be provided to the classifier 130 as the training data I-V_t. The classifier 130 may generate a training model by performing pre-training based on the training data I-V_t. That is, the classifier 130 may perform pre-training based on a plurality of pieces of information about various fault types generated by the simulator 110, and therefore, the reliability of the training model may be improved.

[0040] The classifier 130 can perform a fault analysis operation on the measurement data I-V_m from the measurement device 12 based on the generated training model, and can determine the type of the measurement data I-V_m (e.g., normal state, fault state, or if it is a fault state, determine the fault type). The classifier 130 can output the determined type as a classification result CR.

[0041] The analyzer 140 may perform various operations based on the classification result CR from the classifier 130. For example, the analyzer 140 may be configured to identify the cause of the classified fault based on the classification result CR. Information about the cause of the fault identified by the analyzer 140 may be provided to the manufacturing device 11, and the manufacturing device 11 may eliminate the cause of the fault by controlling various process steps based on the information about the cause of the fault. The above-mentioned operation of the analyzer 140 using the classification result CR is only an example, and example embodiments are not limited thereto.

[0042] As described above, the semiconductor failure analysis device 100 can automatically classify the types of failures occurring in various semiconductor devices included in the semiconductor wafer WF based on the deep neural network algorithm. Therefore, the time spent on analyzing the failures of the semiconductor wafer WF can be reduced. In addition, because the simulator 110 can generate simulation data I-V_s associated with various failure types, and the classifier 130 can perform pre-training based on the simulation data I-V_s, the reliability of the semiconductor failure analysis device 100 can be ensured even if there is no training model for the deep neural network algorithm or the data for pre-training is insufficient.

[0043] Figure 2 1 is a circuit diagram illustrating a memory cell for describing an example embodiment. Below, in order to easily describe the technical features, it is assumed that the semiconductor wafer WF includes a plurality of SRAM devices and the semiconductor failure analysis apparatus 100 performs failure analysis on the SRAM devices. However, example embodiments are not limited thereto.

[0044] For the sake of brevity and ease of description, Figure 2 , one memory cell MC included in the SRAM device is illustrated in FIG. However, example embodiments are not limited thereto, and each of a plurality of SRAM devices included in the semiconductor wafer WF may include a memory cell MC included in the SRAM device. Figure 2 Multiple memory cells similar to the memory cell MC illustrated in FIG.

[0045] Reference Figure 2 , the memory cell MC may include first and second transistors TR1 and TR2 , first and second pull-up transistors PU1 and PU2 , and first and second pull-down transistors PD1 and PD2 .

[0046] The first transistor TR1 is connected between the bit line BL and the first node n1, and the second transistor TR2 is connected between the complementary bit line BLB and the second node n2. The first transistor TR1 and the second transistor TR2 operate based on a signal on the word line WL.

[0047] The first pull-up transistor PU1 has a source terminal connected to the power supply voltage VDD, a drain terminal connected to the first node n1, and a gate terminal connected to the second node n2. The first pull-down transistor PD1 has a source terminal connected to the first node n1, a drain terminal connected to the ground voltage VSS, and a gate terminal connected to the second node n2.

[0048] The source terminal of the second pull-up transistor PU2 is connected to the power supply voltage VDD, the drain terminal is connected to the second node n2, and the gate terminal is connected to the first node n1. The source terminal of the second pull-down transistor PD2 is connected to the second node n2, the drain terminal is connected to the ground voltage VSS, and the gate terminal is connected to the first node n1.

[0049] As described above, the memory cell MC may have a structure of 6 transistors. However, example embodiments are not limited thereto, and the memory cell MC may be implemented in various structures.

[0050] The memory cell MC may include a design for test (DFT) terminal. The measurement device 12 may be configured to measure the measurement data I-V_m through the DFT terminal of the memory cell MC.

[0051] Figure 3 It is an example Figure 1 A flowchart of the operation of the simulator. Figures 1 to 3 In operation S101, the simulator 110 may set variables "i" and "k" to "1". The variables "i" and "k" set in operation S101 are only used to describe the iterative operation of the simulator 110, and example embodiments are not limited thereto.

[0052] In operation S110, the simulator 110 may generate a simulation for the i-th fault type. Due to various process or characteristic problems of the components, a fault may occur in the SRAM device (e.g., Figure 2 Various fault types occur in the memory cell MC of the first node n1, for example, a short circuit between the first node n1 and the ground voltage VSS, a short circuit between the first pull-up transistor PU1 and the power supply voltage VDD, and a short circuit between the first node n1 and the second node n2. The simulator 110 may generate a simulation of a first fault type (because "i" is "1" in operation S101) among the plurality of fault types.

[0053] In operation S120, the simulator 110 may set the kth parameter. For example, various elements (e.g., Figure 2 In the example of FIG. 1 , the transistors TR1, TR2, PU1, PU2, PD1 and PD2, or the resistance, capacitance and inductance of the signal line) can be set to have specific physical values. Since the physical values ​​of various elements are set differently, different simulation results (i.e., simulation data) can be obtained. In this way, the simulator 110 can set the physical value associated with the simulated element as the first parameter (because "k" is "1" in operation S101).

[0054] In operation S130, the simulator 110 may measure electrical data from the generated simulation. For example, the simulator 110 may measure electrical data, such as a current-voltage curve of each element, from the simulation generated by operation S110 and operation S120. In an example embodiment, for one simulation, a plurality of electrical data may be measured.

[0055] In operation S140, the simulator 110 may combine a plurality of measured electrical data to generate simulation data I-V_s. For example, the simulation generated by operation S110 and operation S120 may be a simulation associated with a memory cell MC. A plurality of electrical data (i.e., a plurality of current-voltage data) may be required to classify a fault type associated with a memory cell MC. In other words, a fault type may be classified based on a plurality of current-voltage data. A plurality of current-voltage data measured from a simulation may be combined into a simulation data I-V_s. In this case, a simulation data I-V_s may be a unit data for classifying a fault type associated with a simulation (or a memory cell MC).

[0056] In operation S150, the simulator 110 may determine whether the variable "k" is a maximum value. In the case where the variable "k" is not a maximum value, in operation S160, the variable "k" may be increased by "1", and operations S120 to S150 may be repeatedly performed.

[0057] In the case where the determination result of operation S150 indicates that the variable "k" is the maximum value, in operation S170, the simulator 110 may determine whether the variable "i" is the maximum value. In the case where the variable "i" is not the maximum value, in operation S180, the variable "k" is reset to "1", and the variable "i" is increased by "1" (i.e., i=i+1). Then, the simulator 110 may repeatedly perform operations S110 to S170.

[0058] As described above, by repeatedly performing operations S110 to S180 , the simulator 110 may set a plurality of parameters for each of a plurality of simulations, thereby generating simulation data I-V_s.

[0059] Through the iterative operation of the simulator 110 , simulation data I-V_s associated with each of the various fault types may be obtained, and the obtained simulation data I-V_s may be stored in the fault type database 120 .

[0060] For ease of description, the Figure 3 The operations of the embodiment of the present invention are described in detail, but the exemplary embodiments are not limited thereto. For example, certain operations may be performed simultaneously, or may be performed in a different order than shown.

[0061] FIG. 4A to FIG. 4D is a schematic diagram for describing some of the various types of failures that can occur in a memory cell of an SRAM device. FIG. 5A to FIG. 5D is an example of FIG. 4A to FIG. 4D Schematic diagram of simulation data generated for each fault type from the first to the fourth fault type. FIG. 4A to FIG. 4D , simulations associated with various fault types are illustrated in the form of circuit diagrams.

[0062] like Figure 4A As shown, the first memory cell MC1 may include a first fault FT1. The first fault FT1 may be a fault generated due to a short circuit between the first node n1 and the ground voltage VSS, and is illustrated as a resistor connected between the first node n1 and the ground voltage VSS. The simulator 110 may be based on Figure 4A The illustrated circuit diagram of the first memory cell MC1 generates a simulation associated with a first fault FT1.

[0063] In the generated simulation associated with the first fault FT1, the simulator 110 may set the parameters of each element. For example, even if the same type of fault (e.g., the first fault FT1) occurs, the simulation data I-V_s may change as the value or size of the resistor represented as the first fault FT1 changes. In the generated simulation associated with the first fault FT1, the simulator 110 may measure the simulation data I-V_s while changing the parameters of the element. In this way, even for the same type of fault, a plurality of simulation data I-V_s associated with various parameters may be generated. For example, the simulation data I-V_s may be generated from the simulation associated with the first memory cell MC1 including the first fault FT1. Figure 5A The first simulation data I-V_s1 shown in .

[0064] The first simulation data I-V_s1 may have a form in which various current-voltage curves of the first memory cell MC1 are combined together. Figure 5A In the first simulation data I-V_s1, the first region A1 represents the current ① flowing from the first transistor TR1 to the bit line BL according to the voltage change of the word line WL; the second region A2 represents the current ② flowing from the bit line BL to the first transistor TR1 according to the voltage change of the word line WL; the third region A3 represents the current ③ flowing from the complementary bit line BLB to the second transistor TR2 according to the voltage change of the word line WL; the fourth region A4 represents the current ④ flowing from the second transistor TR2 to the complementary bit line BLB according to the voltage change of the word line WL.

[0065] That is, one simulation data (e.g., the first simulation data I-V_s1) may have a form in which a plurality of current-voltage curves are combined. This is because various current-voltage curves are required to classify a fault type. In an exemplary embodiment, the form of the simulation data I-V_s and the form of the measurement data I-V_m (see Figure 1 ) can be configured to have the same form.

[0066] Figure 4B The second memory cell MC2 including the second fault FT2 is shown. The second fault FT2 may be a fault generated due to a short circuit between the first node n1 and the power supply voltage VDD. Figure 4C A third memory cell MC3 including a third fault FT3 is shown. The third fault FT3 may be a fault generated due to a short circuit between the first node n1 and the second node n2. Figure 4D A fourth memory cell MC4 including a fourth fault FT4 is shown. The fourth fault FT4 may be a fault generated due to a variation in the strength of the first pull-down transistor PD1.

[0067] As reference Figure 4A As described, the simulator 110 can generate Figure 4B The second fault FT2, Figure 4C The third fault FT3 and Figure 4D In the simulation associated with the fourth fault FT4, various parameters may be changed respectively, and the second to fourth simulation data I-V_s2 to I-V_s4 may be generated respectively. In an example embodiment, each of the second to fourth simulation data I-V_s2 to I-V_s4 may be divided into first to fourth areas A1 to A4, and the first to fourth areas A1 to A4 represent the same as the reference Figure 4A and Figure 5A The current-voltage curves associated with the different paths or different elements described.

[0068] In one example embodiment, although a description is given of each of the second to fourth simulation data I-V_s2 to I-V_s4 being divided into the first to fourth areas A1 to A4, example embodiments are not limited thereto. For example, various changes may be made to the number of current-voltage curves included in the simulation data I-V_s.

[0069] In one example embodiment, referring to FIG. 4A to FIG. 4D The first to fourth faults FT1 to FT4 described are exemplary, and there may be fault types other than the first to fourth faults FT1 to FT4 associated with one memory cell. For example, the fault type associated with one memory cell may also include various fault types such as: a fault due to a change in the strength of the pull-up transistors PU1 and PU2 or the pull-down transistors PD1 and PD2, a fault due to a short circuit between the second node n2 and the ground voltage VSS, a fault due to a short circuit between the second node n2 and the power supply voltage VDD, and a fault due to a leakage current of the first node; and the simulator 110 may generate a fault type such as FIG. 5A to FIG. 5D Simulation data associated with each fault type is shown.

[0070] As described above, the simulator 110 may generate a simulation associated with each of the various fault types, and may generate simulation data I-V_s associated with the simulation thus generated. The generated simulation data I-V_s may be used for pre-training of the classifier 130. Therefore, even if there is no training model for the deep neural network algorithm or the data for pre-training is insufficient, the reliability of the semiconductor failure analysis device 100 may be ensured.

[0071] Figure 6 is a block diagram illustrating a semiconductor failure analysis system according to an example embodiment. Figure 6, the semiconductor failure analysis system 20 may include a semiconductor wafer WF, a manufacturing device 21, a measuring device 22, and a semiconductor failure analysis device 200. Figure 1 The semiconductor wafer WF, the manufacturing apparatus 21, and the measuring apparatus 22 are described, and therefore, additional description will be omitted to avoid redundancy.

[0072] The semiconductor failure analysis device 200 can be described as follows Figure 1 In this case, the semiconductor failure analysis device 200 can generate double sampling data I-V_d based on the measurement data I-V_m from the measurement device 22 and the separate reference data I-V_ref, and can perform a failure analysis operation based on the generated double sampling data I-V_d. In this case, the accuracy (i.e., reliability) of the failure type classification operation can be improved.

[0073] In detail, the semiconductor failure analysis apparatus 200 may include a simulator 210 , a failure type database 220 , a classifier 230 , first and second double samplers 240 and 250 , and an analyzer 260 .

[0074] The simulator 210 may generate a simulation associated with a normal memory cell MC, and may generate reference data I-V_ref from the generated simulation. That is, the reference data I-V_ref may be electrical data measured from the normal memory cell MC. The reference data I-V_ref generated from the simulator 210 may be provided to the first double sampler 240 and the second double sampler 250.

[0075] For pre-training of the classifier 230 , the fault type database 220 may include fault data I-V_f associated with various fault types.

[0076] The first double sampler 240 may receive the reference data I-V_ref from the simulator 210, and may receive the fault data I-V_f from the fault type database 220. The first double sampler 240 may output the training data I-V_t based on the received reference data I-V_ref and the received fault data I-V_f. The classifier 230 may generate a training model by performing pre-training based on the training data I-V_t from the first double sampler 240. That is, the training data I-V_t may be data for pre-training of the classifier 230, and may be a difference between the fault data I-V_f and the reference data I-V_ref.

[0077] The second double sampler 250 may output double sampling data I-V_d based on the reference data I-V_ref from the emulator 210 and the measurement data I-V_m from the measurement device 22. In an example embodiment, the double sampling data I-V_d may be a difference between the measurement data I-V_m and the reference data I-V_ref. In an example embodiment, although the first double sampler 240 and the second double sampler 250 are illustrated as separate blocks, example embodiments are not limited thereto. For example, the first double sampler 240 and the second double sampler 250 may be implemented with one functional block.

[0078] The classifier 230 may perform a deep neural network algorithm on the dual sampling data I-V_d from the second dual sampler 250. That is, the classifier 230 may perform a failure analysis operation on the semiconductor wafer WF from which the measurement data I-V_m (i.e., data serving as a basis for the dual sampling data I-V_d) is measured.

[0079] In an example embodiment, various forms of measurement data I-V_m may be obtained according to various fault types of SRAM devices present on the semiconductor wafer WF. In this case, even if the first measurement data and the second measurement data are associated with different fault types, the forms or waveforms of the first measurement data and the second measurement data may be similar. Under this condition, in the case where the classifier can directly perform a fault analysis operation on the first measurement data and the second measurement data, because the first measurement data and the second measurement data are similar, the classifier will classify the first measurement data and the second measurement data as the same fault type.

[0080] Alternatively, due to various factors such as changes in the manufacturing environment and mechanical failures of the measuring device, various noises may exist in the measurement data I-V_m, thereby preventing the classifier from normally performing the failure analysis operation.

[0081] In contrast, according to an exemplary embodiment, the classifier 230 performs an analysis operation on the double sampling data I-V_d based on the measurement data I-V_m and the reference data I-V_ref, rather than only on the measurement data. In this case, as described above, even if the first measurement data and the second measurement data are similar, the first double sampling data based on the first measurement data and the reference data I-V_ref and the second double sampling data based on the second measurement data and the reference data I-V_ref can be clearly distinguished from each other. Therefore, even if similar forms of measurement data are collected with respect to different fault types, the classifier 230 can accurately classify the fault type. That is, in terms of fault analysis by the classifier 230, classification consistency can be improved. According to an exemplary embodiment, the reference data I-V_ref is used as the input data. 8A to 9D More fully describe methods to improve classification consistency.

[0082] Figure 7 It is an example Figure 6 Flow chart of the operation method of the semiconductor failure analysis device. For ease of description, it is assumed that Figure 6 The semiconductor failure analysis apparatus 200 is in a state of generating a training model by performing pre-training based on the training data I-V_t (ie, the difference between the reference data I-V_ref and the failure data I-V_f) from the first double sampler 240.

[0083] Reference Figure 6 and Figure 7 In operation S210, the semiconductor failure analysis apparatus 200 may receive measurement data I-V_m from the measurement apparatus 22, and may generate reference data I-V_ref through the simulator 210. As described above, the measurement data I-V_m may be electrical data measured from the semiconductor wafer WF, and the reference data I-V_ref may be electrical data generated from a simulation generated by the simulator 210 regarding a normal memory cell.

[0084] In operation S220, the semiconductor failure analysis apparatus 200 may generate double sampling data I-V_d based on the measurement data I-V_m and the reference data I-V_ref. For example, the second double sampler 250 of the semiconductor failure analysis apparatus 200 may generate the double sampling data I-V_d by performing double sampling on the measurement data I-V_m and the reference data I-V_ref. That is, the double sampling data I-V_d may be the difference between the measurement data I-V_m and the reference data I-V_ref.

[0085] In operation S230, the semiconductor failure analysis apparatus 200 may perform a failure analysis operation based on the generated dual sampling data I-V_d. For example, as described above, the classifier 230 may generate a training model by performing pre-training based on the training data I-V_t from the first dual sampler 240. The classifier 230 may classify the failure type associated with the dual sampling data I-V_d by performing a deep neural network algorithm on the dual sampling data I-V_d based on the generated training model.

[0086] In operation S240, the semiconductor failure analysis device 200 may output the classification result CR to an external device. For example, the analyzer 260 of the semiconductor failure analysis device 200 may perform various operations based on the classification result CR from the classifier 230. For example, the analyzer 260 may provide the manufacturing device 21 with failure information corresponding to the classification result CR or may notify the manager of the failure information through a separate output device (e.g., a display device). In an example embodiment, the manufacturing device 21 may perform a separate control for eliminating the failure based on the information from the analyzer 260, or the manager who is aware of the failure information may take a separate measure to eliminate the failure.

[0087] FIG. 8A to FIG. 8C Is used to describe Figure 6 A schematic diagram of the operation of a semiconductor failure analysis device. FIG. 8A to FIG. 8C How to improve the consistency of one failure type for a semiconductor failure analysis apparatus 200 according to an example embodiment is described.

[0088] Reference Figure 6 and FIG. 8A to FIG. 8C , the fifth memory cell MC5 may be included in the semiconductor wafer WF. The fifth memory cell MC5 may include a fifth fault FT5. The fifth fault FT5 may be a fault generated due to a short circuit between the bit line BL and the power supply voltage VDD.

[0089] As described above, the measurement data measured by the measuring device 22 may vary as the value of the resistor represented as the fifth fault FT5 varies. Figure 8B As shown in , in the case where the resistor represented as the fifth fault FT5 has a first resistance value, the first measurement data I-V_m1 can be measured; in the case where the resistor represented as the fifth fault FT5 has a second resistance value, the second measurement data I-V_m2 can be measured; and in the case where the resistor represented as the fifth fault FT5 has a third resistance value, the third measurement data I-V_m3 can be measured. In an example embodiment, the measurement data measured from the normal memory cell may be the same as the fourth measurement data I-V_m4.

[0090] like Figure 8B As shown in , some of the first to fourth measurement data I-V_m1 to I-V_m4 may have similar waveforms. For example, the waveform or data value of the third measurement data I-V_m3 may be similar to the waveform or data value of the fourth measurement data I-V_m4. That is, in the case where the classifier 230 can directly perform a fault analysis operation on the measurement data, the classifier 230 may classify both the third measurement data I-V_m3 and the fourth measurement data I-V_m4 as the same fault type or a normal state.

[0091] However, as described above, the third measurement data I-V_m3 is data corresponding to the case where the resistor represented as the fifth fault FT5 has the third resistance value (i.e., measurement data measured from the faulty memory cell), while the fourth measurement data I-V_m4 is measurement data measured from the normal memory cell. That is, in the case where the classifier 230 can directly perform a fault analysis operation on the measurement data, classification consistency may become poor.

[0092] In contrast, according to an example embodiment, the classifier 230 may be configured to Figure 8C The double-sampled data shown in performs analytical operations to improve classification consistency. For example, Figure 8C The first double sampling data I-V_d1 can be Figure 8B a difference between the first measurement data I-V_m1 and the reference data I-V_ref; Figure 8C The second double sampling data I-V_d2 can be Figure 8B a difference between the second measurement data I-V_m2 and the reference data I-V_ref; Figure 8C The third double sampling data I-V_d2 can be Figure 8B The difference between the third measurement data I-V_m2 and the reference data I-V_ref; and Figure 8C The fourth double sampling data I-V_d2 can be Figure 8B The difference between the fourth measurement data I-V_m2 and the reference data I-V_ref.

[0093] like Figure 8C As shown in , the first to fourth double sampling data I-V_d1 to I-V_d4 can be clearly distinguished from each other. For example, the fourth double sampling data I-V_d4 is substantially constant in a plurality of regions A1 to A6, while each of the first to third double sampling data I-V_d1 to I-V_d3 can change in the first region A1 and the fifth region A5. Also, the sizes of the first to fourth double sampling data I-V_d1 to I-V_d4 can be different.

[0094] That is, even if the measurement data associated with a specific fault type and the measurement data associated with a normal state are similar, because the semiconductor fault analysis device 200 according to the example embodiment performs a fault analysis operation by using double sampling data based on the measurement data and the reference data (rather than by using only the measurement data), the classification consistency of the fault type can be improved.

[0095] 9A to 9D Is used to describe Figure 6 A schematic diagram of the operation of a semiconductor failure analysis device. 9A to 9DHow to improve the classification consistency of different failure types for the semiconductor failure analysis apparatus 200 according to an example embodiment is described.

[0096] Reference Figure 6 and 9A to 9D , the semiconductor wafer WF may include a sixth memory cell MC6 having a sixth fault FT6 and a seventh memory cell MC7 having a seventh fault FT7.

[0097] The sixth fault FT6 may indicate a fault generated due to a short circuit between the first node n1 and the word line WL, and the seventh fault FT7 may indicate a fault generated due to a short circuit between the first node n1 and the bit line BL.

[0098] The measuring device 22 can measure and output the sixth measurement data I-V_m6 from the sixth storage unit MC6 (refer to Fig. 9C ), and the seventh measurement data I-V_m7 (reference Fig. 9C ).

[0099] Even if the sixth memory cell MC6 and the seventh memory cell MC7 include different fault types FT6 and FT7, such as Fig. 9C As shown in , the waveform or data value of the sixth measurement data I-V_m6 and the waveform or data value of the seventh measurement data I-V_m7 measured from the sixth storage cell MC6 and the seventh storage cell MC7, respectively, may be similar. That is, as described above, in the case where the classifier can directly perform an analysis operation on the measurement data, the sixth measurement data I-V_m6 and the seventh measurement data I-V_m7 may be classified as the same fault type.

[0100] In contrast, according to an example embodiment, the classifier 230 may classify Fig.9D The sixth double sampling data I-V_d6 and the seventh double sampling data I-V_d7 shown in the figure perform a fault analysis operation. The sixth double sampling data I-V_d6 may be the difference between the sixth measurement data I-V_m6 and the reference data I-V_ref, and the seventh double sampling data I-V_d7 may be the difference between the seventh measurement data I-V_m7 and the reference data I-V_ref.

[0101] like Fig.9DAs shown in , the waveform of the second area A2 and the waveform of the sixth area A6 of the sixth double sampling data I-V_d6 and the waveform of the second area A2 and the waveform of the sixth area A6 of the seventh double sampling data I-V_d7 can be clearly distinguished from each other. That is, even if the sixth measurement data I-V_m6 and the seventh measurement data I-V_m7 have similar forms, because the sixth double sampling data I-V_d6 and the seventh double sampling data I-V_d7 are clearly distinguished from each other, the classifier 230 can classify the sixth measurement data I-V_m6 and the seventh measurement data I-V_m7 into different fault types by performing a fault analysis operation on the sixth double sampling data I-V_d6 and the seventh double sampling data I-V_d7. Therefore, in terms of fault analysis of the classifier 230, classification consistency can be improved.

[0102] Fig.10 is a block diagram illustrating a semiconductor failure analysis system according to an example embodiment. Fig.10 , the semiconductor failure analysis system 30 may include a semiconductor wafer WF, a manufacturing device 31, a measuring device 32, and a semiconductor failure analysis device 300. The semiconductor failure analysis device 300 may include a failure type database 310, a first double sampler 320, a second double sampler 330, a classifier 340, and an analyzer 350. The semiconductor wafer WF, the manufacturing device 31, the failure type database 310, the classifier 340, and the analyzer 350 are described above, and therefore, additional description will be omitted to avoid redundancy.

[0103] Different from Figure 6 The semiconductor failure analysis device 200, Fig.10 The semiconductor failure analysis apparatus 300 may be configured to receive the reference data I-V_ref from the measurement apparatus 32. For example, the semiconductor wafer WF may include a plurality of SRAM devices, and each of the SRAM devices may be in a normal state or in a failure state.

[0104] The semiconductor failure analysis device 300 may determine data corresponding to the normal state through a separate preliminary evaluation from among the measurement data measured by the measurement device 32. The measurement data determined to correspond to the normal state may be used as the reference data I-V_ref.

[0105] The first double sampler 320 and the second double sampler 330 may respectively generate training data I-V_t and double sampling data I-V_d by using reference data I-V_ref from the measurement device 32. The classifier 340 may generate a training model by performing pre-training based on the generated training data I-V_t, and may perform an analysis operation on the generated double sampling data I-V_d.

[0106] Fig.11It is an example Fig.10 A flowchart of the operation of the semiconductor failure analysis device. Fig.10 and Fig.11 In operation S310 , the semiconductor failure analysis apparatus 300 may receive measurement data I-V_m and reference data I-V_ref from the measurement apparatus 32 .

[0107] Then, the semiconductor failure analysis apparatus 300 may perform operations S320 to S340. Operations S320 and S340 may be performed together with Figure 7 Operations S220 to S240 are similar, and thus, additional descriptions will be omitted to avoid redundancy.

[0108] As described above, when the semiconductor failure analysis device 300 receives the reference data I-V_ref from the measurement device 32 and generates the double sampling data I-V_d based on the reference data I-V_ref, errors occurring in the measurement device 32 or noise occurring in the measurement data due to changes in various surrounding environments can be offset in the double sampling data I-V_d. Therefore, the reliability of the semiconductor failure analysis device 300 can be improved.

[0109] Fig.12 is a block diagram illustrating a semiconductor failure analysis system according to an example embodiment. Fig.12 , the semiconductor failure analysis system 40 may include a semiconductor wafer WF, a manufacturing device 41, a measuring device 42, and a semiconductor failure analysis device 400. The semiconductor failure analysis device 400 may include a simulator 410, a failure type database 420, a first double sampler 430 and a second double sampler 440, a classifier 450, and an analyzer 460. Each component of the semiconductor failure analysis system 40 is described above, and therefore, additional description will be omitted to avoid redundancy.

[0110] As in Figure 1 As in the semiconductor failure analysis device 100, Fig.12 The semiconductor failure analysis device 400 can generate various simulation data I-V_s by using the simulator 410. That is, even in the case where data for pre-training is insufficient or there is no training model, the reliability of the semiconductor failure analysis device 400 can be ensured.

[0111] Moreover, as in Figure 6 As in the semiconductor failure analysis device 200, Fig.12 The semiconductor failure analysis apparatus 400 can perform an analysis operation on the double sampling data I-V_d based on the measurement data I-V_m and the reference data I-V_ref instead of only based on the measurement data I-V_m. Therefore, the classification consistency of the semiconductor failure analysis apparatus 400 can be improved.

[0112] Fig.13 is a schematic diagram for describing measurement data according to another example embodiment. Fig.13 The semiconductor wafer WF may include a first memory cell MC1 and a second memory cell MC2 adjacent to the first memory cell MC1. The first memory cell MC1 and the second memory cell MC2 both have Figure 2 The memory cell MC has a similar structure, and therefore, additional description will be omitted to avoid redundancy.

[0113] In one example embodiment, although the measurement data I-V_m, the training data I-V_t, and the simulation data I-V_s described above indicate a current-voltage curve associated with one memory cell MC, example embodiments are not limited thereto.

[0114] For example, in the above example embodiment, the measurement data associated with the first memory cell MC1 can be obtained by controlling the voltage of the word line WL, the voltage of the first bit line BL1, and the voltage of the first complementary bit line BL1B. In contrast, in order to obtain the measurement data associated with the second memory cell MC2, the voltage of the word line WL connected to the second memory cell MC2, the voltage of the second bit line BL2, and the voltage of the second complementary bit line BL2B can be controlled. That is, by driving the second memory cell MC2 (in other words, by driving the signal line connected to the second memory cell MC2), the leakage current flowing out of the first memory cell MC1 can be obtained as the measurement data. In an example embodiment, as described above, the simulator can generate a simulation, and the measurement data can be obtained from the generated simulation.

[0115] The measurement data I-V_m, training data I-V_t, and simulation data I-V_s described with reference to the above exemplary embodiments are examples for ease of description, and the exemplary embodiments are not limited thereto. In addition, the failure types of the SRAM device given in the specific implementation and the accompanying drawings are part of the various failure types that may actually exist in the SRAM device, and the exemplary embodiments are not limited thereto.

[0116] The semiconductor failure analysis device according to an example embodiment may perform an analysis operation (or a semiconductor failure type analysis operation) based on a deep neural network algorithm. In this case, the semiconductor failure analysis device may generate simulation data through simulation associated with various failure types and perform pre-training based on the generated simulation data to generate a training model. Therefore, even in the case where the data for pre-training is insufficient or there is no training model, the analysis result of the semiconductor failure analysis device may be more reliable.

[0117] Alternatively, the semiconductor failure analysis apparatus according to example embodiments may improve classification consistency by performing an analysis operation on dual sampling data based on measurement data and reference data instead of only on the measurement data.

[0118] Fig.14 is a schematic diagram illustrating any one of various semiconductor devices manufactured on a semiconductor wafer according to an example embodiment. An example embodiment in which the semiconductor wafer WF includes an SRAM device is described above, but the example embodiment is not limited thereto. For example, Fig.14 As shown in FIG, the first semiconductor wafer WF1 may include a DRAM device having DRAM cells DRAM. Each DRAM cell DRAM may include an access transistor TR and a storage capacitor C, and may be connected to a corresponding word line WL and a corresponding bit line BL.

[0119] In one example embodiment, the semiconductor failure analysis apparatus may perform failure analysis on DRAM devices included in the first semiconductor wafer WF1 based on the above example embodiments.

[0120] For example, the semiconductor failure analysis apparatus may generate a training model by generating simulations associated with various types of failures that may occur in a DRAM device and performing pre-training based on training data from the generated simulations. Alternatively, the semiconductor failure analysis apparatus may generate dual sampling data based on electrical data measured from a DRAM device and separate reference data, and may perform a failure analysis operation on the generated dual sampling data. In this way, the reliability of the semiconductor failure analysis apparatus may be improved.

[0121] Fig.15 is a schematic diagram illustrating any one of various semiconductor devices manufactured on a semiconductor wafer according to an example embodiment. The above describes an example embodiment in which the semiconductor wafer WF includes an SRAM device or a DRAM device, but the example embodiment is not limited thereto. For example, Fig.15 As shown in , the second semiconductor wafer WF2 may include a NAND flash memory device having a flash memory.

[0122] For example, Fig.15 As shown in , the second semiconductor wafer WF2 may include a NAND flash memory block. The NAND flash memory block may include a plurality of cell strings CS11, CS12, CS21, and CS22. The plurality of cell strings CS11, CS12, CS21, and CS22 may be arranged along row directions and column directions to form rows and columns.

[0123] Each of the plurality of cell strings CS11, CS12, CS21, and CS22 includes a plurality of cell transistors. For example, each of the cell strings CS11, CS12, CS21, and CS22 may include string selection transistors SSTa and SSTb, a plurality of memory cells MC1 to MC8, ground selection transistors GSTa and GSTb, and dummy memory cells DMC1 and DMC2. In an example embodiment, each of the plurality of cell transistors included in the cell strings CS11, CS12, CS21, and CS22 may be a charge trap flash (CTF) memory cell.

[0124] In each cell string, a plurality of memory cells MC1 to MC8 may be connected in series and may be stacked in a direction perpendicular to a plane defined by a row direction and a column direction (i.e., in a height direction). In each cell string, string selection transistors SSTa and SSTb may be connected in series and may be inserted between memory cells MC1 to MC8 and a corresponding one of bit lines BL1 and BL2. Ground selection transistors GSTa and GSTb may be connected in series and may be inserted between memory cells MC1 to MC8 and a common source line CSL.

[0125] In an example embodiment, in each cell string, the first dummy memory cell DMC1 may be inserted between the memory cells MC1 to MC8 and the ground selection transistors GSTa and GSTb. In an example embodiment, in each cell string, the second dummy memory cell DMC2 may be inserted between the memory cells MC1 to MC8 and the string selection transistors SSTa and SSTb.

[0126] The ground selection transistors GSTa and GSTb of the cell strings CS11, CS12, CS21, and CS22 may be connected to the ground selection line GSL in common. In an example embodiment, the ground selection transistors in the same row may be connected to the same ground selection line, and the ground selection transistors in different rows may be connected to different ground selection lines. For example, the first ground selection transistors GSTa of the cell strings CS11 and CS12 in the first row may be connected to the first ground selection line, and the first ground selection transistors GSTa of the cell strings CS21 and CS22 in the second row may be connected to the second ground selection line.

[0127] Memory cells at the same height from the substrate or ground selection transistors GSTa and GSTb are connected to the same word line, and memory cells at different heights from the substrate or ground selection transistors GSTa and GSTb are connected to different word lines. For example, memory cells MC1 to MC8 of cell strings CS11, CS12, CS21, and CS22 may be connected to the first to eighth word lines WL1 to WL8.

[0128] The string selection transistors belonging to the same row among the first string selection transistors SSTa of the same height are connected to the same string selection line, and the string selection transistors belonging to different rows are connected to different string selection lines. For example, the first string selection transistor SSTa of the cell string CS11 in the first row and the first string selection transistor SSTa of the cell string CS12 can be connected to the string selection line SSL1a in common, and the first string selection transistor SSTa of the cell string CS21 in the second row and the first string selection transistor SSTa of the cell string CS22 can be connected to the string selection line SSL2a in common.

[0129] Similarly, string selection transistors belonging to the same row among the second string selection transistors SSTb of the same height can be connected to the same string selection line, and string selection transistors in different rows can be connected to different string selection lines. For example, the second string selection transistor SSTb of the cell string CS11 in the first row and the second string selection transistor SSTb of the cell string CS12 can be connected to the string selection line SSL1b in common, and the second string selection transistor SSTb of the cell string CS21 in the second row and the second string selection transistor SSTb of the cell string CS22 can be connected to the string selection line SSL2b in common.

[0130] In an exemplary embodiment, dummy memory cells of the same height are connected to the same dummy word line, and dummy memory cells of different heights are connected to different dummy word lines. For example, the first dummy memory cell DMC1 is connected to the first dummy word line DWL1, and the second dummy memory cell DMC2 is connected to the second dummy word line DWL2.

[0131] As described above, patterns for various semiconductor devices such as DRAM devices, SRAM devices, and flash memory devices may be formed on a semiconductor wafer WF. However, example embodiments are not limited thereto. A semiconductor failure analysis apparatus according to an example embodiment may perform failure analysis operations on various semiconductor elements or various semiconductor devices.

[0132] According to example embodiments, a semiconductor failure analysis apparatus may execute a deep neural network algorithm to automatically classify various types of failures occurring on a semiconductor wafer. Even in the absence of a training model for a failure analysis operation or insufficient data for pre-training, the semiconductor failure analysis apparatus according to example embodiments may obtain various simulation data for pre-training through simulation.

[0133] Alternatively, the semiconductor failure analysis apparatus according to example embodiments may improve the consistency of failure analysis operations by performing failure analysis operations on double sampling data based on measurement data and reference data. Therefore, a semiconductor failure analysis apparatus having improved reliability and a failure analysis method thereof are provided.

[0134] The various operations of the above-described methods may be performed by any suitable means capable of performing the operations (eg, various (one or more) hardware and / or software components, circuits, and / or (one or more) modules).

[0135] The software may include an ordered arrangement of executable instructions for implementing logical functions and may be contained in any "processor-readable medium" for use by an instruction execution system, apparatus or device (e.g., a single-core or multi-core processor or a system containing a processor) or in conjunction with an instruction execution system, apparatus or device.

[0136] The blocks or steps or functions of the methods or algorithms described in conjunction with the example embodiments disclosed herein may be directly embodied as hardware, software modules executed by a processor, or a combination of the two. If implemented in software, the functions may be stored as one or more instructions or codes on or transmitted through a tangible, non-transitory computer-readable medium. The software module may be located in a random access memory (RAM), a flash memory, a read-only memory (ROM), an electrically programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), a register, a hard disk, a removable disk, a CD ROM, or any other form of storage medium known in the art.

[0137] As conventional, the example embodiments are described in the form of functional blocks, units and / or modules and illustrated in the accompanying drawings. It will be appreciated by those skilled in the art that these blocks, units and / or modules are physically implemented by electronic (or optical) circuits such as logic circuits, discrete components, microprocessors, hard-wired circuits, storage elements, wiring connections, etc., wherein these circuits can be formed using semiconductor-based manufacturing techniques or other manufacturing techniques. In the case where blocks, units and / or modules are implemented by microprocessors or similar devices, software (e.g., microcode) can be used to program blocks, units and / or modules to perform various functions discussed herein, and blocks, units and / or modules can be optionally driven by firmware and / or software. Alternatively, each block, unit and / or module can be implemented by dedicated hardware, or as a combination of dedicated hardware that performs certain functions and processors (e.g., one or more programmed microprocessors and related circuits) that perform other functions. Moreover, without departing from the scope of the present disclosure, each block, unit and / or module of the example embodiment can be physically divided into two or more interactive and discrete blocks, units and / or modules. Furthermore, the blocks, units and / or modules of the example embodiments may be physically combined into more complex blocks, units and / or modules without departing from the scope of the present disclosure.

[0138] Although example embodiments have been described, it will be apparent to those of ordinary skill in the art that various changes and modifications can be made thereto without departing from the spirit and scope of the present disclosure as set forth in the following claims.

Claims

1. A failure analysis method for a semiconductor failure analysis device, the failure analysis method include: receiving measurement data measured from semiconductor devices included in a semiconductor wafer; generating dual sampling data based on the measurement data and the reference data; performing a fault analysis operation on the double sampling data; classifying a failure type of the semiconductor device based on a result of the failure analysis operation; as well as Output information about the fault type. in, The fault analysis method further comprises: generating training data based on the reference data and failure data associated with each of a plurality of failure types of the semiconductor device; and Performing pre-training based on the training data to generate a training model for a deep neural network algorithm, Wherein, the fault analysis operation is performed based on the training model for the deep neural network algorithm.

2. The fault analysis method according to claim 1, in, The reference data is data associated with at least one normal current-voltage curve corresponding to a normal state of the semiconductor device, and Wherein, the measurement data is data associated with at least one measured current-voltage curve measured from the semiconductor device.

3. The fault analysis method according to claim 1, in, The double sampling data corresponds to a difference between the measurement data and the reference data.

4. The fault analysis method according to claim 1, in, The training data corresponds to the difference between the fault data and the reference data.

5. The fault analysis method according to claim 1, in, The fault data is generated from a simulation generated for each of the plurality of fault types.

6. The fault analysis method according to claim 5, in, The simulation is generated based on an integrated circuit simulation program.

7. The fault analysis method according to claim 1, in, The measurement data is measured through a design for testability terminal of the semiconductor device.

8. The fault analysis method according to claim 1, in, The semiconductor device is a static random access memory.

9. A semiconductor failure analysis device, include: a measuring device configured to measure electrical data from a semiconductor device included in a semiconductor wafer and output the measured electrical data as measurement data; a first double sampler configured to generate double sampling data based on the measurement data and reference data; a classifier configured to perform a failure analysis operation based on the dual sampling data and classify a failure type of the semiconductor device based on a result of the failure analysis operation; as well as an analyzer, the analyzer being configured to perform analysis based on the fault type and output a result of the analysis, in, The semiconductor failure analysis device further comprises: a memory having a failure type database including failure data associated with a plurality of failure types of the semiconductor device; and a second double sampler configured to generate training data based on the fault data and the reference data, The classifier is further configured to generate a training model for performing the fault analysis operation based on the training data. Wherein, the fault analysis operation is performed based on the training model for the deep neural network algorithm.

10. The semiconductor failure analysis device according to claim 9, in, The double sampling data corresponds to a difference between the measurement data and the reference data.

11. The semiconductor failure analysis device according to claim 9, further comprising: include: a simulator configured to generate a simulation for each of the plurality of failure types of the semiconductor device and to measure simulation data from the simulation, The simulation data is stored in the fault type database as the fault data.

12. The semiconductor failure analysis device according to claim 11, in, The simulator is based on an integrated circuit simulator.

13. The semiconductor failure analysis device according to claim 11, in, The simulator is also configured to: generating a normal simulation associated with a normal state of the semiconductor device; and The reference data is generated based on the normal simulation.

14. The semiconductor failure analysis device according to claim 9, in, The reference data is output from the measuring device.

15. A semiconductor failure analysis device, include: a measurement device configured to receive electrical data from a semiconductor device included in a semiconductor wafer and output the electrical data as measurement data; a simulator configured to generate a simulation for each of a plurality of failure types of the semiconductor device and to generate simulation data based on the simulation; a classifier configured to generate a training model by performing pre-training based on the simulation data, perform a failure analysis operation on the measurement data by using the training model, and classify a failure type of the semiconductor device based on the failure analysis operation; as well as An analyzer is configured to perform analysis based on the fault type and output a result of the analysis.

16. The semiconductor failure analysis device according to claim 15, in, The simulator is based on an integrated circuit simulator.

17. The semiconductor failure analysis device according to claim 15, in, The measurement data is data associated with at least one measured current-voltage curve measured from the semiconductor device, and The simulation data is data associated with at least one simulated current-voltage curve measured from the simulation.