Method and system for testing gas composition inside a microcavity of a MEMS wafer level package

Abstract

This invention relates to the field of microelectromechanical system (MEMS) packaging and testing, and discloses a method and system for testing the gas composition inside a microcavity in a MEMS wafer-level package. The method includes the following steps: acquiring the coordinate information and geometric parameters of a target microcavity on the wafer under test; aligning a locally sealed sampling head with the surface of the region where the target microcavity is located; releasing gas from the target microcavity into the locally sealed sampling cavity; updating the posterior probability distribution of the gas component concentration vector in real time based on a Bayesian inference framework; inputting the enriched sample gas into a residual gas analyzer; and retrieving the gas composition and concentration inside a single microcavity based on gas type and partial pressure information. Precise positioning is achieved by acquiring the coordinate information and geometric parameters of the target microcavity, and adaptive optimal stopping of the accumulated number is achieved by updating the posterior probability distribution of the gas component concentration vector in real time during the sequential opening process based on a Bayesian inference framework.

Description

Technical Field

[0001] This invention relates to the field of microelectromechanical system (MEMS) packaging and testing technology, specifically to a method and system for testing the gas composition inside the microcavities of MEMS wafer-level packaging. Background Technology

[0002] In the wafer-level packaging process, microelectromechanical systems (MEMS) devices form sealed microcavities. The type and concentration of residual gases within these cavities directly affect the device's quality factor, frequency stability, and long-term reliability. Typical wafer-level vacuum-packaged or hermetically sealed MEMS devices include gyroscopes, resonators, microbolometers, pressure sensors, and optical MEMS switches. As devices evolve towards higher performance and longer lifespans, relying solely on changes in total pressure within the cavity to assess package hermeticity is no longer sufficient for engineering requirements.

[0003] Existing methods for testing the gas composition inside packaged microcavities mostly involve opening the packaged device after it has been discretely packaged at the device level, or puncturing it with a screw, and then introducing the released gas into an analysis chamber for mass spectrometry detection. However, the geometric dimensions of a single MEMS wafer-level packaged microcavity are typically only a few hundred micrometers square and tens of micrometers high, with a single cavity volume as low as nanoliters. The absolute mass of a target component inside is far below the detection limit of conventional residual gas analyzers. Only a single puncture sampling is performed on a single package. When the gas volume in a single cavity is insufficient, an effective spectrum cannot be obtained, which can easily lead to detection failure or require multiple repeated sampling, resulting in low testing efficiency. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for testing the gas composition inside microcavities of MEMS wafer-level packaging. This solves the problems of existing methods for testing the gas composition inside microcavities of packaging, which are prone to detection failure or require multiple repeated sampling, resulting in low testing efficiency.

[0005] To achieve the above objectives, the present invention provides a method for testing the gas composition inside a microcavity of a MEMS wafer-level package, comprising the following steps: The coordinate information and geometric parameters of the target microcavity on the wafer under test are obtained. The target microcavity is the product working cavity or the witness microcavity formed by the same packaging process as the product working cavity. The local micro-sealed sampling head is brought into contact with the surface of the area where the target microcavity is located to form a local sampling cavity, and the local sampling cavity and its connecting passage are evacuated. A release opening is formed in the puncture guide area of ​​the target microcavity to release the gas inside the target microcavity into the local sampling cavity; When the amount of gas released is lower than the preset analysis threshold, the witness microcavity is opened sequentially, and the released gas is accumulated and enriched. During the sequential opening of the witness microcavities, the posterior probability distribution of the gas component concentration vector is updated in real time based on the Bayesian inference framework, and subsequent witness microcavities are opened based on the comparison between the uncertainty measure of the posterior probability distribution and the preset acceptable uncertainty upper limit. Once the preset stop conditions are met, the enriched sample gas is input into the residual gas analyzer to obtain gas type and partial pressure information; The composition and concentration of gas inside a single microcavity can be retrieved based on the type of gas and its partial pressure.

[0006] This invention provides a method and system for testing the gas composition inside the microcavity of a MEMS wafer-level package. It offers the following advantages: 1. This invention achieves precise positioning by acquiring the coordinate information and geometric parameters of the target microcavity. A low dead volume sampling cavity is formed by bonding a locally micro-sealed sampling head with the wafer surface. After evacuation, a release opening is formed by puncturing the guide area. By sequentially opening the witness microcavities and accumulating and enriching the released gas, the limitation of insufficient gas volume in a single cavity is overcome. During the sequential opening process, the posterior probability distribution of the gas component concentration vector is updated in real time based on a Bayesian inference framework. The decision on whether to continue opening subsequent witness microcavities is dynamically made based on the comparison between the uncertainty metric and the preset acceptable upper limit. This achieves adaptive optimal stopping of the accumulated number of microcavities, solving the problem that existing methods for testing the gas composition inside encapsulated microcavities are prone to detection failure or require multiple repeated sampling, resulting in low testing efficiency.

[0007] 2. This invention introduces a Bayesian inference framework during the sequential opening of witness microcavities, updates the posterior probability distribution of the gas component concentration vector in real time, and dynamically determines whether to continue opening subsequent witness microcavities based on the comparison between the uncertainty measure of the posterior probability distribution and the preset acceptable upper limit of uncertainty. This achieves adaptive optimal stopping of the accumulated enrichment quantity and can automatically adjust the accumulation depth according to the amount of information contained in the actual measurement data, avoiding detection failure or excessive accumulation caused by fluctuations in sample gas volume, and significantly improving testing efficiency and the rationality of sample utilization.

[0008] 3. This invention achieves a successive recursive update of the posterior probability distribution by pre-setting the prior distribution of the gas component concentration vector and fusing the observation data obtained from opening each witness microcavity with the prior distribution according to Bayes' formula. This allows prior knowledge in historical batch data or packaging process parameters to be effectively utilized, improving the accuracy of concentration inference under small sample conditions. At the same time, the output form of the posterior probability distribution provides a complete probabilistic description of the concentration estimation, including the posterior mean and confidence interval, which quantifies the uncertainty of the test results, making it easier for engineers to assess the reliability of the data and make engineering decisions accordingly.

[0009] 4. In the process of sequentially opening the witness microcavities, in addition to adopting a preset order, this invention can also prioritize the opening of candidate microcavities with the largest expected information gain based on the principle of maximizing mutual information, further reducing the number of opening cavities required to reach the preset confidence level. It makes full use of existing observation data to predict the amount of information in untested microcavities, optimizes the sampling order, and inputs the enriched sample gas into the residual gas analyzer in the form of pulses. Combined with the standard gas calibration coefficient, blank spectrum subtraction results, and single cavity volume and number of opening cavities, the concentration is inverted, and the posterior mean and confidence interval of the concentration of each component are output. While ensuring the accuracy of the test results, it provides quantifiable confidence information. Attached Figure Description

[0010] Figure 1 This is a flowchart of the method for testing the gas composition inside the microcavity of MEMS wafer-level packaging proposed in this invention; Figure 2 This is an architecture diagram of the testing system for the gas composition inside the microcavity of MEMS wafer-level packaging proposed in this invention; Figure 3 This is a planar-sectional view of the witness microcavity array and the product working cavity provided in an embodiment of the present invention; Figure 4 A cross-sectional view of the partially micro-sealed sampling head and the puncture guide area provided in an embodiment of the present invention; Figure 5 This is a timing diagram of sequential cumulative sampling and pulse injection provided in an embodiment of the present invention; Figure 6 The simulation results of the characteristic partial pressure varying with the number of accumulated cavities and dead volume are provided in the embodiments of the present invention. Figure 7 The simulation results of detection confidence and concentration inversion error provided in the embodiments of the present invention are shown in the figure. Figure 8 The simulation results are shown in the figure for the relationship between the orifice diameter, release efficiency, and pressure transients provided in the embodiments of the present invention. Figure 9 The simulation results of the evolution of characteristic gases under different failure modes are shown in the embodiments of the present invention. Detailed Implementation

[0011] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0012] Example 1: In a first embodiment of the present invention, the present invention provides a method for testing the gas composition inside a microcavity of a MEMS wafer-level package, such as... Figure 1 As shown, it includes the following steps: Obtain the coordinate information and geometric parameters of the target microcavity on the wafer under test. The target microcavity is the product working cavity or a witness microcavity formed using the same packaging process as the product working cavity. Further, the coordinate information and geometric parameters of the target microcavity on the wafer under test are obtained, including the following steps: By fusing at least two types of information from microscopic images, infrared images, map coordinates, and grazing marks through a visual positioning and coordinate mapping module, the puncture coordinates with an alignment error of no more than 5 micrometers are output. The length, width, and height of the microcavity are obtained through layout design or 3D reconstruction, and the geometric volume of a single microcavity is calculated.

[0013] Specifically, obtaining the coordinate information and geometric parameters of the target microcavity on the wafer under test is to provide basic data for the accurate alignment of the subsequent local micro-sealed sampling head and the inversion of single-cavity concentration. The wafer under test simultaneously contains the product working cavity and the witness microcavity formed by the same packaging process as the product working cavity. The witness microcavity is set in the dicing lane, the monitoring chip area or the independent witness area. High-precision coordinate positioning is achieved by fusing multi-source visual information, and the volume parameters of a single microcavity are obtained through geometric modeling.

[0014] In this embodiment, the precise location of the target microcavity is obtained through a visual positioning and coordinate mapping module. Simultaneously, at least two types of information are acquired from microscopic images, infrared images, layout coordinates, and scribe line markings. Let the pixel coordinates of the i-th feature point extracted from the microscopic image be denoted as the image coordinate vector, and its corresponding wafer physical coordinates be given by the layout design values ​​as the world coordinate vector. A mapping relationship is established through an affine transformation model. , where the transformation matrix It is a 2×2 rotation and scaling matrix. For a 2×1 translation vector, after solving the transformation parameters using at least three feature points, the alignment error of the output piercing coordinates is preferably no greater than 5 micrometers.

[0015] The volume of a single cavity is calculated using a cuboid approximation model, with the length of the microcavity set as follows: Width is The height is The volume formula is: This geometric parameter will be fed into the concentration inversion module for use in subsequent formulas. The dilution factor is calculated in the following way: To observe the partial pressure, To improve overall transmission efficiency, For the local sampling cavity dead volume, For the original partial pressure, accurately obtain This is a prerequisite for ensuring the accuracy of the inversion.

[0016] The local micro-sealed sampling head is made to fit against the surface of the target microcavity area to form a local sampling cavity, and the local sampling cavity and its connecting passage are evacuated. Furthermore, the local micro-sealed sampling head is brought into contact with the surface of the target microcavity area to form a local sampling cavity, and the local sampling cavity and its connecting passage are evacuated, including the following steps: The annular sealing lip of the local micro-sealed sampling head conformally fits to the wafer surface under the control of clamping force feedback, forming a local sampling cavity with a dead volume of 5 nanoliters to 2 microliters. The pressure in the local sampling chamber and connecting pipeline is reduced to no higher than 1×10 using a turbomolecular pump assembly. - ³Pascal.

[0017] Specifically, after obtaining the precise coordinates and geometric volume of the target microcavity, the local micro-sealed sampling head needs to be moved above the coordinates and form a sealed space with the wafer surface. At the same time, the space is evacuated to a vacuum state. This step creates a sampling environment with low background interference for subsequent gas release and avoids ambient gas from mixing into the sample gas. The dead volume of the local sampling cavity directly determines the degree of dilution of the sample gas. The smaller the dead volume, the higher the detection sensitivity. However, an excessively small dead volume will increase the sealing difficulty and evacuation time.

[0018] The annular sealing lip of the local micro-seal sampling head conformally fits to the wafer surface under pressure feedback control. An integrated force sensor monitors the contact pressure between the sealing lip and the wafer surface in real time. Once the sampling head descends to a predetermined position, a servo motor or pneumatic actuator applies a clamping force, maintaining its value within a preset range. The sealing lip material is low-outgassing fluororubber or polytetrafluoroethylene, with an outgassing rate of less than 1 × 10⁻⁶ at room temperature. -8 Pascals per second; in this embodiment, the dead volume of the local sampling cavity. The dead volume is set in the range of 5 nanoliters to 2 microliters, preferably 20 nanoliters to 200 nanoliters, and mainly consists of the internal volume of the sampling head cavity and the internal volume of the connecting pipeline.

[0019] After sealing is achieved, the local sampling chamber and its connecting passages are evacuated using a turbomolecular pump assembly. The forestage of the turbomolecular pump is provided with a rough vacuum by a rotary vane pump or diaphragm pump, ultimately reducing the pressure within the local sampling chamber to a high vacuum state. During the evacuation process, a full-range vacuum gauge is used to monitor pressure changes in real time. In this embodiment, the target evacuation pressure is set to be no higher than 1 × 10⁻⁶. - ³ Pascal, preferably 8 × 10 -4Pascal, evacuation time is generally 90 to 180 seconds, depending on the dead volume and tubing admittance; after evacuation, the isolation valve is closed to maintain the vacuum state of the local sampling chamber. At this time, the system is in the open-hole state. This evacuation operation simultaneously collects blank background spectrum, which is used for background subtraction in subsequent data processing.

[0020] Observed partial pressure after gas release Compared with the original partial pressure within the microcavity The relationship between them is described by the following formula: ,in, The overall transmission efficiency of gas from the microcavity to the residual gas analyzer is 0.6 to 0.9, preferably 0.72. The geometric volume of a single microcavity. The dead volume of the local sampling cavity, under the same original partial pressure Down, The smaller the value, the better the observed partial pressure. The larger the value, the higher the detection sensitivity; therefore, Controlling the process within the optimal window is crucial for ensuring test success. This step, by creating a low dead volume and high vacuum sampling environment, provides reliable physical conditions for the subsequent sequential opening of the witness microcavity and cumulative enrichment.

[0021] A release opening is formed in the puncture guide area of ​​the target microcavity, allowing the gas inside the target microcavity to be released into the local sampling cavity; Furthermore, a release opening is formed in the puncture guide area of ​​the target microcavity to release the gas inside the target microcavity into the local sampling cavity, including the following steps: By employing microneedle puncture, femtosecond laser micro-aperture, or ion beam aperture methods, release openings with an equivalent diameter of 1 to 25 micrometers are formed in the puncture guidance area; The success of the opening was confirmed by detecting the rate of change of pressure in the sampling chamber using a local pressure transient monitoring unit.

[0022] Specifically, after the local sampling chamber is in a vacuum state, the cap of the target microcavity needs to be opened in a controllable manner to release the gas inside. The guide area is punctured as a pre-designed thinned area or brittle covering layer, with a thickness of 10% to 80% of the thickness of the adjacent sealing area. This is intended to reduce the fluctuation of the opening force and suppress the generation of debris. Under the premise of ensuring the integrity of the microcavity structure, a small opening is formed to allow the gas inside the cavity to be released quickly into the local sampling chamber, while avoiding debris from clogging the pipeline or contaminating the enrichment unit.

[0023] A release opening is formed in the puncture guide area using microneedle piercing, femtosecond laser micro-aperture, or ion beam aperture methods. When microneedle piercing is selected, a hollow microneedle with an outer diameter of 8-15 micrometers is used. Under controlled pressure, it is vertically inserted into the puncture guide area, and the needle tip retracts after penetrating the cap, forming a circular opening with an equivalent diameter of approximately 5-12 micrometers. When femtosecond laser aperture is selected, a laser beam with a pulse width of less than 300 femtoseconds and a wavelength of 515 nanometers is used, with the single pulse energy controlled between 0.5-2 microjoules. The opening is etched by point-by-point scanning or spiral scanning. In this embodiment, the equivalent diameter of the release opening is... The setting is 1-25 micrometers, preferably 3-12 micrometers, which controls the risk of particulate contamination caused by the opening while ensuring the gas release rate.

[0024] After successful orifice opening, the gas inside the microcavity rapidly flows into the local sampling chamber under pressure differential. The gas flow rate is related to the orifice diameter, the internal pressure, and the downstream vacuum. To confirm the success of the orifice opening, the pressure change rate in the sampling chamber is detected by a local pressure transient monitoring unit. This unit uses a high-precision capacitive pressure gauge with a sampling frequency of no less than 1 kHz. At the moment of orifice opening, the pressure signal... It exhibits step-up characteristics, and its rate of ascent satisfies the following relationship: ,in, This represents the rate of pressure change in the initial stage after the hole is opened. For flow conduction of the opening, and the aperture The cube is approximately proportional to it, that is... ; This refers to the dead volume of the local sampling cavity; The pressure difference between the microcavity pressure and the initial pressure of the local sampling cavity is used to determine whether the hole opening is successful when the monitored pressure change rate exceeds the preset threshold; if the change rate is too low or there is no change, the hole opening fails, the system issues an alarm signal and stops subsequent operations. This transient confirmation method can provide real-time feedback on the hole opening quality and avoid invalid accumulation.

[0025] When the amount of gas released is lower than the preset analysis threshold, the witness microcavity is opened sequentially, and the released gas is accumulated and enriched. Furthermore, when the amount of released gas is lower than a preset analysis threshold, the witness microcavity is sequentially activated, and the released gas is accumulated and enriched, including the following steps: Each witness micro-cavity is opened sequentially according to a preset order or based on the principle of maximizing mutual information. Each released gas is introduced into a low-dead-volume pipeline into a cryogenic cold trap, an adsorbent micro-enrichment trap, or a micro-volume storage cavity for accumulation until a preset stopping condition is reached.

[0026] Specifically, when the amount of gas released from a single microcavity is initially assessed to be below the detection limit of the residual gas analyzer, multiple witness microcavities with the same process need to be opened sequentially, and the gas released each time needs to be accumulated and enriched to increase the total amount of sample gas to meet the analysis requirements. The preset analysis threshold is determined comprehensively based on the sensitivity of the residual gas analyzer, the dead volume of the local sampling cavity, and the transmission efficiency of the target gas. This step overcomes the limitation of insufficient gas volume in a single cavity through accumulation and enrichment, and at the same time provides multiple sets of observation data for subsequent Bayesian adaptive decision-making.

[0027] The witness microcavities can be sequentially activated either using a preset order or based on the principle of maximizing mutual information. When using a preset order, the microcavities are activated sequentially from left to right and from top to bottom according to their physical location on the wafer. When using the principle of maximizing mutual information, the activation is based on the currently acquired observation data. The corresponding posterior probability distribution Calculate each candidate witness microcavity The expected mutual information is determined, and the cavity with the highest mutual information is preferentially selected for opening. The formula for estimating mutual information is: ,in, Indicating existing observations Below, to be observed With unknown concentration Mutual information between them; For Shannon's differential entropy, Indicates the prediction distribution The expected value is determined by approximating the mutual information of each candidate cavity using the Monte Carlo method, and then selecting the cavity with the largest mutual information value as the next cavity to be opened. If the mutual information of all candidate cavities is lower than a preset threshold of 10, the cavity is not opened. - If ³Nate, the accumulation process will be terminated prematurely.

[0028] Each time the witness microcavity is opened, the released gas is introduced into the enrichment unit through a low dead volume pipeline for accumulation. The enrichment unit can be a cryogenic cold trap, an adsorbent micro-enrichment trap, or a micro-volume storage cavity. Before entering the enrichment unit, each released gas passes through a capillary with a dead volume of less than 10 nanoliters to minimize sample gas loss. The accumulation process continues until a preset stop condition is reached. This stop condition includes, but is not limited to, the number of accumulated microcavities reaching a preset upper limit, the current accumulated sample gas being estimated to meet the detection requirements, or the mutual information being lower than a threshold. Through the above accumulation and enrichment steps, the nanoliter-level single-cavity gas that was originally undetectable is enriched to a detectable level, providing sufficient sample gas for subsequent pulse injection and mass spectrometry analysis.

[0029] During the sequential opening of the witness microcavities, the posterior probability distribution of the gas component concentration vector is updated in real time based on the Bayesian inference framework, and subsequent witness microcavities are opened based on the comparison between the uncertainty measure of the posterior probability distribution and the preset acceptable uncertainty upper limit. Furthermore, during the sequential opening of the witness microcavity, the posterior probability distribution of the gas component concentration vector is updated in real time based on a Bayesian inference framework, including the following steps: The prior distribution of the gas component concentration vector is set based on the historical test data or packaging process parameters of this batch of wafers; After each witness microcavity is opened, the obtained observation data is fused with the prior distribution according to Bayes' theorem to calculate the updated posterior probability distribution.

[0030] Furthermore, based on the comparison between the uncertainty measure of the posterior probability distribution and the preset acceptable upper limit of uncertainty, the subsequent witness microcavity is activated, including the following steps: Calculate the uncertainty measure of the current posterior probability distribution. If the uncertainty measure is greater than the preset acceptable uncertainty upper limit, continue to open the next witness microcavity. Accumulation and enrichment stop when the uncertainty measure is less than or equal to the preset acceptable upper limit of uncertainty.

[0031] Specifically, when witness microcavities are opened sequentially and the released gas is accumulated and enriched, the observation data obtained from each opened microcavity contains information about the concentration of the real gas components. In order to make full use of this information and dynamically decide when to stop accumulation, a Bayesian inference framework is introduced. The unknown concentration is treated as a random variable, and the knowledge state before the test is expressed by the prior distribution. Then, the observation data and the prior distribution are fused by the Bayesian formula to continuously update the belief about the concentration, i.e., the posterior probability distribution. Subsequently, based on the comparison between the uncertainty measure of the posterior distribution and the preset acceptable upper limit of uncertainty, it is adaptively determined whether to continue opening subsequent witness microcavities. This step integrates Bayesian inference in probability statistics with engineering decision-making, realizing the leap from a fixed threshold to adaptive optimal stopping.

[0032] First, based on historical test data or packaging process parameters of this batch of wafers, a priori distribution of the gas component concentration vector is established. Wafers in the same batch use the same packaging process and getter configuration, and their microcavity gas concentrations have similar distribution characteristics. Let the gas component concentration vector be... ,in To represent the number of target gas components, the prior distribution can be a multivariate normal distribution or a log-normal distribution. Taking the partial pressure of nitrogen as an example, let its prior distribution be... ,in This is the historical average of the batch. Let be the standard deviation within the batch. If there is no historical data, then the maximum entropy no-information prior is used, for example, let . , Choose a larger value to represent weak information.

[0033] After each witness microcavity is opened, the acquired observation data is fused with the prior distribution according to Bayes' theorem to calculate the updated posterior probability distribution. Let the... The observed partial pressure of the witness microcavity is This observation is consistent with the actual concentration. The two sides satisfy a linear Gaussian model: ,in, To improve overall transmission efficiency, For the geometric volume of a single microcavity, For the local sampling cavity dead volume, To measure the noise variance, the prior distribution is normally distributed. When the posterior distribution is still a normal distribution, its update formula is: ; ; For example, let the initial prior... =38 Pascals Pascal, Nassen, Nassen, Pascal, Observations Pascal, then the updated posterior mean Approximately 39.2 Pascals, posterior standard deviation The value is approximately 6 Pascals, significantly reducing uncertainty; the calculation process is repeated once for each microcavity opened, resulting in the real-time evolution of concentration beliefs.

[0034] The decision to continue activating subsequent witness microcavities is based on a comparison between the uncertainty measure of the posterior probability distribution and the preset acceptable upper limit of uncertainty. It can be defined as the posterior standard deviation, posterior variance, or safety criticality risk function, using safety criticality risk: That is, the posterior probability that the true partial pressure exceeds a certain safety threshold, with a preset upper limit for acceptable uncertainty. Set according to project requirements, such as requirements The specific judgment logic is as follows: Calculate the uncertainty measure of the current posterior probability distribution. When the uncertainty measure is greater than the preset acceptable upper limit of uncertainty, it indicates that the current cognitive accuracy is insufficient to meet engineering requirements, and the system continues to open the next witness microcavity; when the uncertainty measure is less than or equal to the preset acceptable upper limit of uncertainty, it indicates that sufficiently reliable information has been obtained, and the system stops accumulating and enriching. In this embodiment, taking the posterior standard deviation as an example, let the acceptable upper limit be 0.8 Pascals. When the calculated... The comparison stops when the Pascal is reached; through this adaptive comparison mechanism, a preset confidence level can be achieved with the minimum number of cavities opened.

[0035] Once the preset stop conditions are met, the enriched sample gas is input into the residual gas analyzer to obtain gas type and partial pressure information; Furthermore, once the preset stop conditions are met, the enriched sample gas is input into the residual gas analyzer to obtain gas type and partial pressure information, including the following steps: Thermal desorption is performed on the enrichment unit used for accumulation and enrichment, so that the accumulated sample gas is sent into the residual gas analyzer in the form of a pulse valve. Set the mass-to-charge ratio scanning range, collect the ion current intensity of the target gas component, convert the partial pressure value of each gas component according to the preset correspondence between ion current intensity and partial pressure, and identify the gas type based on the mass-to-charge ratio.

[0036] Specifically, once the preset stopping condition is met through Bayesian adaptive decision-making, sufficient sample gas has been collected in the accumulation enrichment unit. It needs to be released and sent to the residual gas analyzer for mass spectrometry detection. This step converts the enriched sample gas into quantifiable gas types and partial pressure information, providing basic data for subsequent concentration inversion. The residual gas analyzer uses a quadrupole mass spectrometer, whose detection principle is to separate the ions according to their mass-to-charge ratio after electron bombardment ionization and measure the intensity of the ion current.

[0037] First, the enrichment unit used for cumulative enrichment undergoes thermal desorption. When the enrichment unit is an adsorbent micro-enrichment trap, the temperature is rapidly raised to 150-250 degrees Celsius by a heating resistance wire integrated on its outer wall, and held for 30-60 seconds, causing the adsorbed gas molecules to desorb rapidly from the adsorbent surface. When the enrichment unit is a low-temperature cold trap, the cooling source is removed while heating is performed simultaneously, causing the condensed gas to vaporize. The desorbed sample gas is propelled by the carrier gas helium and flows through a low dead volume transfer line to the pulse valve. The pulse valve opens for a few milliseconds under the controller's command, forming a narrow pulse sample gas cloud, which is sent into the analysis chamber of the residual gas analyzer in the form of a pulse with a time width of less than 50 milliseconds. The pulse injection mode can improve the instantaneous concentration and enhance the signal-to-noise ratio.

[0038] Set the mass-to-charge ratio scan range and collect the ion current intensity of the target gas component. The control software of the residual gas analyzer has a preset scan range of 1-100 atomic mass units, covering hydrogen, water vapor, nitrogen, oxygen, argon, carbon dioxide, and common organic debris. The residence time at each mass-to-charge ratio position is approximately 10-50 milliseconds, with continuous cyclic scanning. Ion current intensity partial pressure of corresponding gas components There is a linear relationship between them: ,in, For mass-to-charge ratio indication, This represents the partial pressure of the gas corresponding to that mass-to-charge ratio. The ion current intensity measured by the detector. This is the sensitivity coefficient. The background baseline offset is used; the sensitivity coefficient and baseline offset are calibrated by pre-testing standard gas or reference microcavity. At the same time, the gas type can be identified by comparing the mass-to-charge ratio value with the standard spectral library. For example, a mass-to-charge ratio of 18 corresponds to water vapor, and a mass-to-charge ratio of 44 corresponds to carbon dioxide. Through this step, the partial pressure value of each gas component and its corresponding type identifier are output. These data will be sent to the concentration inversion module to convert the original concentration of the single cavity.

[0039] The composition and concentration of gas inside a single microcavity can be retrieved based on the type of gas and its partial pressure.

[0040] Furthermore, based on the gas type and partial pressure information, the gas composition and concentration inside a single microcavity are retrieved, including the following steps: The single-cavity volume in the geometric volume, the number of witness microcavities actually opened during the accumulation and enrichment process, the blank spectrum subtraction result obtained by pre-opening the unsealed reference microcavity or collecting the blank spectrum of the local sampling cavity, and the standard gas calibration coefficient obtained by opening the reference microcavity containing the standard gas. The original partial pressure of each gas component in a single microcavity is calculated based on the preset gas transmission efficiency. Output the posterior mean and confidence interval of each component concentration.

[0041] Specifically, after obtaining the partial pressure values ​​of each mass-charge ratio output by the residual gas analyzer, it is necessary to convert them into the original gas partial pressure in a single microcavity. This conversion process needs to comprehensively consider factors such as the cumulative number of witness microcavities opened, the volume ratio of local sampling chambers to microcavities, gas transmission efficiency, and background interference subtraction. This step restores the measured values ​​to concentration data that can reflect the real packaging atmosphere, providing a quantitative basis for failure determination and process evaluation.

[0042] First, obtain the volume of the single cavity in the geometric volume. The number of witness microcavities actually opened during the accumulation and enrichment process The blank spectrum subtraction result and the standard gas calibration coefficient are included. Derived from the aforementioned steps of obtaining geometric parameters, The blank spectrum subtraction result was obtained by recording the Bayesian adaptive stopping decision process. This was achieved by pre-opening an unsealed reference microcavity or acquiring the background spectrum of a local sampling cavity before opening the aperture. The standard gas calibration coefficient was obtained by opening a reference microcavity containing a known concentration of standard gas and calibrating the residual gas analyzer response. Indicates the first The proportional relationship between the measured partial pressure of a gas component and its original partial pressure.

[0043] Calculate the original partial pressure of each gas component in a single microcavity according to the preset gas transport efficiency. Let the first... The measured partial pressure of the gas components is The original partial pressure of this component within a single microcavity Calculated by the following formula: ,in, The output of the residual gas analyzer partial pressure of the gas, This is the background partial pressure corresponding to the blank spectrum, used to subtract pipeline memory effect and system background; The actual number of witness microcavities opened, dimensionless. The overall transmission efficiency of gas from the microcavity to the residual gas analyzer is preferably 0.72; It is the reciprocal of the dilution factor. This is the standard gas calibration coefficient, dimensionless, and ideally 1.

[0044] Output the posterior mean and confidence interval of each component concentration, since the concentration has already been obtained during the aforementioned Bayesian inference process. Based on the posterior probability distribution, this step can extract posterior samples from it, calculate the mean as a point estimate, and take the 2.5% and 97.5% quantiles as the 95% confidence interval; for example, the posterior mean of nitrogen concentration is 40.1 Pascals, and the 95% confidence interval is [38.8, 41.4] Pascals. This output not only gives the best estimate but also quantifies the uncertainty of the estimate, making it easier for engineers to judge the reliability of the data. Through this step, the complete test process starts from the original measurement signal and finally obtains statistically significant single-cavity gas composition and concentration data.

[0045] Example 2: In a second embodiment of the present invention, the present invention provides a testing system for the gas composition inside the microcavity of a MEMS wafer-level package, such as... Figure 2 As shown, it includes the following modules: Coordinate acquisition module: used to acquire the coordinate information and geometric parameters of the target microcavity on the wafer under test. The target microcavity is the product working cavity or a witness microcavity formed using the same packaging process as the product working cavity. Sealing and evacuation module: used to make the local micro-sealed sampling head fit against the surface of the target micro-cavity area to form a local sampling cavity and to evacuate the local sampling cavity and its connecting passages; Opening and gas release module: used to form a release opening in the puncture guide area of ​​the target microcavity, so that the gas in the target microcavity can be released into the local sampling cavity; Accumulation and enrichment module: When the amount of released gas is lower than the preset analysis threshold, the witness microcavity is opened sequentially and the released gas is accumulated and enriched. Decision module: Used to update the posterior probability distribution of the gas component concentration vector in real time based on the Bayesian inference framework during the sequential opening of witness microcavities, and to open subsequent witness microcavities based on the comparison between the uncertainty measure of the posterior probability distribution and the preset acceptable uncertainty upper limit. Sample injection analysis module: used to input the enriched sample gas into the residual gas analyzer after the preset stop conditions are met, and obtain gas type and partial pressure information; Concentration Inversion Module: Used to invert the gas composition and concentration inside a single microcavity based on gas type and partial pressure information.

[0046] A MEMS wafer foundry needed to test the internal gas composition of microcavities in a batch of gyroscope wafers that had completed wafer-level vacuum packaging to evaluate the hermeticity of the packaging and the activity of the getter. This batch consisted of 24 wafers, each with over 2000 working cavities. The dicing area had 36 witness microcavities whose process was completely synchronized with the working cavities. Traditional methods required destructive sampling of individual packages after wafer dicing, which was inefficient and failed to capture the process uniformity information of the entire wafer. To solve these problems, the MEMS wafer-level package microcavity internal gas composition testing system provided by this invention was adopted, the architecture of which is as follows: Figure 2 As shown. The specific implementation process of this system is as follows: First, the coordinate acquisition module reads the wafer layout file, integrates the microscopic image with the dicing marks, and outputs the puncture coordinates of each witness microcavity. The alignment error is controlled within 3 micrometers, and the geometric volume of a single microcavity is calculated to be 2.25 nanoliters. The sealing and evacuation module controls the local micro-sealed sampling head to move above the first witness microcavity. The annular sealing lip adheres to the wafer surface under a clamping force of 1.2 Newtons, forming a local sampling cavity with a dead volume of approximately 80 nanoliters. The turbomolecular pump assembly pumps the pressure inside the cavity to 8 × 10⁻⁶. -4 Pascal; The opening and gas release module uses a hollow microneedle with an equivalent diameter of 8 micrometers to puncture the puncture guide area, and the pressure transient monitoring unit confirms that the opening is successful. The accumulation and enrichment module sequentially activates the subsequent witness microcavities, and the gas released each time is introduced into the adsorbent micro-enrichment trap through the low dead volume pipeline for accumulation. After each microcavity is opened, the decision module updates the posterior probability distribution according to the prior distribution set by the historical batch data and the current observation data using the Bayesian formula, calculates the posterior standard deviation of the nitrogen partial pressure, and stops accumulating when the posterior standard deviation drops to 0.65 Pascal, which is lower than the preset acceptable upper limit of 0.8 Pascal, and actually opens 6 witness microcavities. The sample injection analysis module performs thermal desorption of the enrichment unit at 200 degrees Celsius, and then pulses the sample into the quadrupole residual gas analyzer. It scans the mass-to-charge ratio range of 1 to 100 atomic mass units, and collects the ion current intensity of nitrogen, argon, water vapor and carbon dioxide and converts it into partial pressure values. Based on the single-cavity volume of 2.25 nanoliters, the number of open cavities of 6, the blank spectrum subtraction results, and the standard gas calibration coefficient, the concentration inversion module calculated the original nitrogen partial pressure in a single microcavity to be 40.1 Pascals, with a 95% confidence interval of 38.8 Pascals to 41.4 Pascals, according to a transmission efficiency of 0.72. Meanwhile, no abnormalities were found in the water vapor partial pressure, indicating that the packaging atmosphere of this batch was qualified and the getter was working properly.

[0047] Example 3: The testing system provided in this embodiment consists of a wafer carrier and vacuum isolation module, a visual positioning and coordinate mapping module, a local micro-sealed sampling head, a micro-aperture execution module, an enrichment trap and pulse valve, a vacuum evacuation module, an RGA spectrum acquisition module, and a concentration inversion and failure cause determination module, forming a complete link from positioning, release, accumulation, sample injection to cause determination.

[0048] The wafer carrier and vacuum isolation module is used to carry the wafer under test and maintain a local sampling environment. It includes an adjustable wafer stage and a local vacuum isolation cavity. The visual positioning and coordinate mapping module is used to output the positioning coordinates of the target microcavity and the puncture guide area. It integrates at least two types of information from microscopic images, infrared images, layout coordinates, and scribe marks, and outputs puncture coordinates with an alignment error of no more than 5 μm.

[0049] The localized micro-sealed sampling head includes a sampling head body, an annular sealing lip, and a clamping force feedback unit. The sealing material is a low-outgassing material conformally fitted to the wafer surface, used to create a repeatedly openable sealed sampling space for a localized area of ​​the target microcavity without covering the entire wafer. The micro-aperture execution module is used to create a release opening in the puncture guide area, and the opening method includes microneedle puncture, laser micro-aperture, ion beam aperture, or a combination thereof.

[0050] The gas enrichment and pulsed injection module includes an enrichment trap and a pulse valve. Enrichment methods include cryogenic cold traps, adsorbent micro-enrichment traps, micro-volume storage cavities, or combinations thereof. The residual gas analysis and concentration inversion module includes an RGA spectrum acquisition module and a concentration inversion and failure cause determination module, used to output gas type, partial pressure, and single-cavity concentration results.

[0051] The test procedure provided in this embodiment includes: reading coordinates and parameters, calibrating the reference cavity and establishing the baseline, local evacuation, forming a release opening in the puncture guide area, determining whether the signal-to-noise ratio meets the standard, if not, continuing to open the subsequent witness microcavity, accumulating sample gas in the enrichment unit, performing pulse injection when the accumulated amount reaches the threshold, acquiring RGA spectrum, and determining concentration inversion and failure mode.

[0052] like Figure 3 As shown, a product working cavity array 302 and a witness microcavity array 303 are disposed on wafer 301. The witness microcavity array 303 is located in the dicing area, spatially separated from the product working cavity array 302 but maintaining process synchronization. Positioning marks 304 are used to establish coordinate mapping and visual recognition relationships. The cross-sectional view shows the longitudinal structural relationship between the capping layer, the puncture guide area, the product working cavity, the getter / sensitive material, and the witness microcavities.

[0053] like Figure 4 As shown, the cross-sectional structure of the locally micro-sealed sampling head includes a sampling head body 401, a locally sampling cavity 402 formed by the sampling head body and the wafer surface, a sealing lip 403, a capping layer 404, a piercing guide area 405, and a microneedle or laser focus 406. In operation, the sealing lip 403 adheres to the wafer surface to form a seal, and the microneedle or laser focus 406 pierces the guide area 405 to form a release opening, allowing gas from the product working cavity 407 or witness microcavity 408 to be released into the locally sampling cavity 402.

[0054] like Figure 5 As shown, the timing relationship between sequential cumulative sampling and pulse injection is as follows: First, the reference cavity is calibrated to establish a baseline, and then the witness microcavity is opened and gas is released sequentially. The released gas continues to accumulate in the enrichment trap. When the accumulated amount reaches the preset threshold, the pulse valve is opened to inject the sample, and the residual gas analyzer begins to collect spectral data.

[0055] The key process parameter windows in this embodiment are as follows: the feasible window for the dead volume of the local sampling chamber is 5 nL to 2 μL, and the preferred window is 20 nL to 200 nL, to reduce background dilution while ensuring sealing stability; the feasible window for the equivalent diameter of the puncture opening is 1 μm to 25 μm, and the preferred window is 3 μm to 12 μm, to balance release efficiency, debris control, and sealing integrity; the feasible window for the cumulative number of witness microcavities is 2 to 200, and the preferred window is 10 to 60, to address insufficient gas volume per cavity while ensuring testing efficiency; the puncture alignment error is preferably no greater than 5 μm to avoid accidental and off-center punctures; the local evacuation base pressure is preferably no higher than 1 × 10⁻⁶. - ³Pa, to reduce background gas and memory effect.

[0056] To verify the rationality of the above parameter window, an independent simulation tool was used to conduct engineering-level simulation analysis on the key parameters. The simulation model was based on an ideal gas mixing model, a logical confidence model, an aperture-risk integrated model, a pressure transient ordinary differential equation model, and a concentration inversion error Monte Carlo model. The basic simulation parameters were set as follows: the single-chamber volume was calculated to be approximately 2.25 nL based on a diameter of 300 μm × 300 μm × 25 μm; the single-chamber total pressure was set to 40 Pa; the target component volume fraction was set to 18%; the overall transport efficiency was set to 72%; the background baseline was set to 0.06 Pa; and the reliable partial pressure threshold was set to 0.45 Pa.

[0057] like Figure 6 As shown, the simulation results reveal the relationship between the characteristic partial pressure and the cumulative number of witness cavities under different dead volume conditions. The results indicate that under an 80 nL dead volume condition, activating 4 witness microcavities is sufficient to achieve a reliable partial pressure threshold of 0.45 Pa; under a 200 nL dead volume condition, approximately 12 witness microcavities need to be activated to achieve a detection confidence level of over 90%.

[0058] like Figure 7 As shown, the heatmaps for detection confidence and concentration inversion error indicate that, within the overlapping region of the high confidence and low error areas, the combination of dead volume and the number of witness microcavities constitutes the recommended optimal design window. Under conditions of 12 witness cavities and 0.06 Pa noise, the median relative error is 2.67%.

[0059] like Figure 8 As shown, the relationship between orifice diameter and release efficiency, overall risk score, and pressure transients indicates that the optimal orifice diameter for the overall score is approximately 5.5 μm, falling within the preferred window of 3 μm to 12 μm. With an 8 μm orifice diameter, the time to reach 95% steady-state release is approximately 3.51 ms, which can serve as a rapid confirmation window for successful orifice opening.

[0060] like Figure 9 As shown, the evolution curves of characteristic gases with aging time under three typical failure modes indicate that: in the leakage mode, N2, O2 and Ar increase simultaneously; in the release mode, H2O, CO2 and organic fragment peaks increase preferentially; and in the getter failure mode, H2 and previously adsorbed active gases are released again.

[0061] This embodiment performs cumulative testing of a witness cavity array on a wafer-level vacuum-packaged gyroscope sample. Thirty-six witness microcavities are arranged in a 6x6 array in the scribe line region. Each witness microcavity has a planar dimension of approximately 300 μm x 300 μm and a cavity height of approximately 25 μm. The capping material uses a silicon-silicon bonding structure, the bonding material is Au-Sn eutectic solder, the getter is a Ti-based non-evaporable getter film, and the encapsulation atmosphere is high-purity N2. All the above parameters are consistent with those of the product's working cavity.

[0062] Witness cavity array accumulation mode: Step 1: The positioning module reads the positioning marks on the scribe line and the coordinates of the witness microcavity array to establish a puncture coordinate mapping. The positioning module integrates optical microscopic images and layout coordinate information, and the alignment error is controlled within 3μm.

[0063] Step 2: Lower the local micro-sealed sampling head above the target witness microcavity area, ensuring the sealing lip adheres to the wafer surface to form a local sampling cavity of approximately 80 nL. The sealing material is low-emission fluororubber, and the clamping force is controlled between 0.5 N and 2 N via force sensor feedback.

[0064] Step 3: Evacuate the local sampling chamber and connecting pipeline using a turbomolecular pump assembly, reducing the base pressure to 8 × 10⁻⁶. - 4 Pa, the evacuation time is approximately 120 seconds.

[0065] Step 4: Using a hollow microneedle with an equivalent aperture diameter of 8 μm, a release opening is formed in the puncture guide area of ​​the first witness microcavity under controlled pressure. The cap thickness of the puncture guide area is approximately 30% of the thickness of the adjacent sealing area. After opening, the success of the opening is confirmed by a local pressure transient monitoring unit.

[0066] Step 5: Perform the same opening operation on the 2nd to 12th witness microcavities in the preset order. After each opening, the gas released is introduced into the adsorption enrichment unit through the low dead volume pipeline for accumulation.

[0067] Step 6: After all 12 witness microcavities are opened, thermal desorption is performed on the enrichment unit at a temperature of approximately 200°C for approximately 30 seconds. The accumulated sample gas is then sent into the quadrupole mass spectrometer residual gas analyzer via a pulse valve.

[0068] Step 7: The residual gas analyzer detected four main spectral peaks: N2 (mass-to-charge ratio 28), Ar (mass-to-charge ratio 40), H2O (mass-to-charge ratio 18), and CO2 (mass-to-charge ratio 44). Combining the single-cavity volume of 2.25 nL, the number of open witness cavities of 12, the blank spectrum subtraction results, and the standard gas calibration coefficients, the equivalent gas concentration distribution of a single microcavity was obtained by inversion.

[0069] Test results show that N2 is the main gaseous component in this batch of samples, the Ar concentration is consistent with the residual amount in the packaging atmosphere, the H2O background is slightly higher than expected but does not exceed the upper limit of process control, and no obvious O2 peak is detected, indicating that there is no obvious external gas leakage in this batch of samples, but there is a slight background of water vapor release.

[0070] Direct verification mode for product working cavity: This embodiment demonstrates direct verification of the working chamber of a wafer-level hermetically sealed infrared microcalorimeter sample. First, a witness microcavity with the same manufacturing process is selected to calibrate the response coefficient of the residual gas analyzer. During calibration, two reference microcavities containing a mixture of standard gases of known concentrations are opened, and the released gas is fed into the residual gas analyzer to obtain the response coefficients of each component.

[0071] Subsequently, a localized release was performed on the working cavity of one product in the failure analysis sample using a laser micro-aperture method. A femtosecond laser was used with a pulse width of approximately 200 fs, a wavelength of 515 nm, and a single pulse energy of approximately 1 μJ, creating an opening with an equivalent diameter of approximately 6 μm in the piercing guide area. Two reference witness microcavities were simultaneously activated as blanks and response controls.

[0072] Analysis results show that the signal intensities of H2O and organic debris peaks in the product's working cavity are significantly higher than those in the reference witness microcavity, while the ratio of N2 and Ar is basically consistent with the packaging atmosphere. Based on this, it is determined that the increased cavity pressure in the failed sample is more likely due to the release of gas from the sealing material and adhesive layer after packaging, rather than external gas leakage. This implementation method is suitable for small-scale destructive verification of R&D samples, small-sample failure analysis, or project evaluation scenarios.

[0073] Accelerated aging cause analysis mode: This embodiment focuses on accelerated aging causal analysis of MEMS resonator samples with getter. Wafers from the same batch were divided into group A and group B. Group A underwent high-temperature storage testing at 125°C for 168h, 500h, and 1000h, respectively. Group B underwent temperature cycling testing within a temperature range of -40°C to 125°C for 200, 500, and 1000 cycles, respectively. At each time point, samples were removed, and sequential gas release accumulation analysis was performed on the witness microcavity array. In each test, 15 witness microcavities were activated for accumulation.

[0074] The results of the high-temperature storage test in Group A showed that the H2O and CO2 peaks continued to rise over time, the organic fragment peaks increased significantly after 500 hours, while the N2 and Ar peaks remained relatively stable. Based on this, it was determined that the increase in cavity pressure under high-temperature storage conditions mainly originated from the thermal release of gas from the encapsulation material.

[0075] The results of the temperature cycling test in Group B showed that after 500 cycles, the peaks for N2, O2, and Ar began to rise synchronously, and this upward trend accelerated in subsequent cycles. Based on this, it was determined that temperature cycling may have led to the formation of microcracks at the bonding interface, causing a slow leakage of external gas. The H2 peak showed a rebound trend after 1000 cycles, suggesting that the getter may be beginning to show signs of insufficient capacity or decreased activity.

[0076] This implementation method allows for the further decomposition of total pressure changes into contributions from different gas components, thereby providing direct quantitative evidence for process improvement.

[0077] This invention can form at least the following four typical application scenarios: Scenario 1 is direct verification of the product working cavity, which is suitable for small-scale sample verification and failure analysis project identification in the R&D stage; Scenario 2 is witness microcavity array accumulation, which is suitable for batch-level packaging atmosphere evaluation in the mass production introduction stage; Scenario 3 is reliability aging tracking, which is suitable for sampling and analyzing witness microcavities at different time points in accelerated aging tests; Scenario 4 is incoming material and packaging verification, which is suitable for incoming material inspection and packaging process verification scenarios.

[0078] Systematic testing was conducted on 18 MEMS wafers from three batches, 72 witness microcavity arrays, and 24 product working cavity verification points. The control scheme was a traditional destructive baseline scheme involving direct puncture of a single cavity without enrichment. Key verification data are as follows: the effective spectral recognition rate increased from 34% in the baseline scheme to 97% in this scheme; the relative standard deviation of single-cavity concentration repeatability decreased from 27% to 6%; the localization hit rate increased from 62% to 96%; and the single-wafer completion time was shortened from 180 min to 48 min.

[0079] This embodiment also provides a MEMS wafer-level packaging witness structure, including multiple witness microcavities disposed in the dicing lane, monitoring chip area or independent witness area. The witness microcavities and the product working cavity maintain process synchronization in terms of sealing material, cavity height, bonding interface and packaging atmosphere, and are gas-isolated from each other. Each witness microcavity is provided with a positioning mark and a puncture guide area above it so that the local micro-sealed sampling head can sample the witness microcavities sequentially.

[0080] The puncture guide zone consists of a thinned zone, a brittle capping layer, a prefabricated micropore plugging layer, or a combination thereof, to reduce pore force fluctuations and suppress debris from entering the local sampling cavity.

[0081] The number of witness microcavities ranges from 2 to 200, preferably from 10 to 60, and they are distributed in row and column arrays or ring arrays in the dicing channel or monitoring chip area.

[0082] The witness structure also includes a reference microcavity for calibrating the response coefficient of the residual gas analyzer, which contains a standard gas or a mixture of standard gases of known types and concentrations.

[0083] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for testing the gas composition inside a microcavity of a MEMS wafer-level package, characterized in that, Includes the following steps: The coordinate information and geometric parameters of the target microcavity on the wafer under test are obtained. The target microcavity is the product working cavity or the witness microcavity formed by the same packaging process as the product working cavity. The local micro-sealed sampling head is brought into contact with the surface of the area where the target microcavity is located to form a local sampling cavity, and the local sampling cavity and its connecting passage are evacuated. A release opening is formed in the puncture guide area of ​​the target microcavity to release the gas inside the target microcavity into the local sampling cavity; When the amount of gas released is lower than the preset analysis threshold, the witness microcavity is opened sequentially, and the released gas is accumulated and enriched. During the sequential opening of the witness microcavities, the posterior probability distribution of the gas component concentration vector is updated in real time based on the Bayesian inference framework, and subsequent witness microcavities are opened based on the comparison between the uncertainty measure of the posterior probability distribution and the preset acceptable uncertainty upper limit. Once the preset stop conditions are met, the enriched sample gas is input into the residual gas analyzer to obtain gas type and partial pressure information; The composition and concentration of gas inside a single microcavity can be retrieved based on the type of gas and its partial pressure.

2. The method for testing the gas composition inside a MEMS wafer-level packaged microcavity according to claim 1, characterized in that: The process of obtaining the coordinate information and geometric parameters of the target microcavity on the wafer under test includes the following steps: By fusing at least two types of information from microscopic images, infrared images, map coordinates, and grazing marks through a visual positioning and coordinate mapping module, the puncture coordinates with an alignment error of no more than 5 micrometers are output. The length, width, and height of the microcavity are obtained through layout design or 3D reconstruction, and the geometric volume of a single microcavity is calculated.

3. The method for testing the gas composition inside a MEMS wafer-level packaged microcavity according to claim 1, characterized in that: The process of attaching the locally sealed sampling head to the surface of the target microcavity region to form a local sampling cavity and evacuating the local sampling cavity and its connecting passages includes the following steps: The annular sealing lip of the local micro-sealed sampling head conformally fits to the wafer surface under the control of clamping force feedback, forming a local sampling cavity with a dead volume of 5 nanoliters to 2 microliters. The pressure in the local sampling chamber and connecting pipeline is reduced to no higher than 1×10 using a turbomolecular pump assembly. - ³Pascal.

4. The method for testing the gas composition inside a MEMS wafer-level packaged microcavity according to claim 1, characterized in that: The process of forming a release opening in the puncture guide area of ​​the target microcavity to release gas from the target microcavity into the local sampling cavity includes the following steps: By employing microneedle puncture, femtosecond laser micro-aperture, or ion beam aperture methods, release openings with an equivalent diameter of 1 to 25 micrometers are formed in the puncture guidance area; The success of the opening was confirmed by detecting the rate of change of pressure in the sampling chamber using a local pressure transient monitoring unit.

5. The method for testing the gas composition inside a MEMS wafer-level packaged microcavity according to claim 1, characterized in that: The process of sequentially opening the witness microcavity, and updating the posterior probability distribution of the gas component concentration vector in real time based on a Bayesian inference framework, includes the following steps: The prior distribution of the gas component concentration vector is set based on the historical test data or packaging process parameters of this batch of wafers; After each witness microcavity is opened, the obtained observation data is fused with the prior distribution according to Bayes' theorem to calculate the updated posterior probability distribution.

6. The method for testing the gas composition inside a MEMS wafer-level packaged microcavity according to claim 1, characterized in that: The step of activating the subsequent witness microcavity by comparing the uncertainty measure of the posterior probability distribution with a preset acceptable upper limit of uncertainty includes the following steps: Calculate the uncertainty measure of the current posterior probability distribution, and continue to open the next witness microcavity when the uncertainty measure is greater than the preset acceptable uncertainty upper limit; Accumulation and enrichment stop when the uncertainty measure is less than or equal to a preset acceptable upper limit of uncertainty.

7. The method for testing the gas composition inside a MEMS wafer-level packaged microcavity according to claim 1, characterized in that: When the amount of released gas is lower than a preset analysis threshold, the witness microcavity is sequentially opened, and the released gas is accumulated and enriched, including the following steps: Each witness micro-cavity is opened sequentially according to a preset order or based on the principle of maximizing mutual information. Each released gas is introduced into a low-dead-volume pipeline into a cryogenic cold trap, an adsorbent micro-enrichment trap, or a micro-volume storage cavity for accumulation until a preset stopping condition is reached.

8. The method for testing the gas composition inside a MEMS wafer-level packaged microcavity according to claim 1, characterized in that: Once the preset stopping condition is met, the enriched sample gas is input into the residual gas analyzer to obtain gas type and partial pressure information, including the following steps: Thermal desorption is performed on the enrichment unit used for accumulation and enrichment, so that the accumulated sample gas is sent into the residual gas analyzer in the form of a pulse valve. Set the mass-to-charge ratio scanning range, collect the ion current intensity of the target gas component, convert the partial pressure value of each gas component according to the preset correspondence between ion current intensity and partial pressure, and identify the gas type based on the mass-to-charge ratio.

9. The method for testing the gas composition inside a MEMS wafer-level packaged microcavity according to claim 1, characterized in that: The process of retrieving the gas composition and concentration inside a single microcavity based on gas type and partial pressure information includes the following steps: The single-cavity volume in the geometric volume, the number of witness microcavities actually opened during the accumulation and enrichment process, the blank spectrum subtraction result obtained by pre-opening the unsealed reference microcavity or collecting the blank spectrum of the local sampling cavity, and the standard gas calibration coefficient obtained by opening the reference microcavity containing the standard gas. The original partial pressure of each gas component in a single microcavity is calculated based on the preset gas transmission efficiency. Output the posterior mean and confidence interval of each component concentration.

10. A testing system for the gas composition inside a microcavity of a MEMS wafer-level package, characterized in that: The method applied to any one of claims 1-9 includes: Coordinate acquisition module: used to acquire the coordinate information and geometric parameters of the target microcavity on the wafer under test, wherein the target microcavity is the product working cavity or the witness microcavity formed by the same packaging process as the product working cavity; Sealing and evacuation module: used to make the local micro-sealed sampling head fit against the surface of the target micro-cavity area to form a local sampling cavity and evacuate the local sampling cavity and its connecting passage; Opening and gas release module: used to form a release opening in the puncture guide area of ​​the target microcavity, so that the gas in the target microcavity is released into the local sampling cavity; Accumulation and enrichment module: When the amount of released gas is lower than the preset analysis threshold, the witness microcavity is opened sequentially and the released gas is accumulated and enriched. Decision module: used to update the posterior probability distribution of the gas component concentration vector in real time based on the Bayesian inference framework during the sequential opening of the witness microcavities, and to open the subsequent witness microcavities based on the comparison between the uncertainty measure of the posterior probability distribution and the preset acceptable uncertainty upper limit. Sample injection analysis module: used to input the enriched sample gas into the residual gas analyzer after the preset stop conditions are met, and obtain gas type and partial pressure information; Concentration Inversion Module: Used to invert the gas composition and concentration inside a single microcavity based on gas type and partial pressure information.

Images