12-inch silicon wafer double-sided polishing upper and lower surface removal amount monitoring system and method

Abstract

The invention relates to the technical field of semiconductor material precision machining, and discloses a 12-inch silicon wafer double-sided polishing upper and lower surface removal amount monitoring system and method, and the system comprises a macroscopic thickness monitoring unit, a microcosmic contact state decoupling sensing unit and a data processing controller. The method comprises the following steps: synchronously acquiring a total thickness signal of a silicon wafer, an acoustic emission signal of an upper / lower fixed disc side and a fluid dynamic pressure signal; the method comprises the following steps: constructing a lubrication state correction coefficient by utilizing fluid dynamic pressure, and carrying out nonlinear correction on a mechanical friction energy index extracted by acoustic emission to obtain an effective vector removal index representing real cutting capacity; and the removal distribution ratio of the upper surface and the lower surface is calculated, and the independent absolute removal amount of the upper surface and the lower surface is obtained through decoupling in combination with the total thickness variation. According to the method, by eliminating the interference of a fluid'water cushion effect 'on a friction signal, visualization and closed-loop feedback control of a single-face removal behavior in the double-face polishing process are achieved, and the problem of silicon wafer warping caused by removal asymmetry is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of precision processing technology for semiconductor materials, specifically to a system and method for monitoring the amount of material removed from the top and bottom surfaces during double-sided polishing of a 12-inch silicon wafer. Background Technology

[0002] As the feature size of integrated circuits continues to shrink, the semiconductor industry has placed extremely high demands on the surface flatness and geometry of 12-inch large-size single-crystal silicon wafers. Double-side polishing (DSP), a key process in silicon wafer processing that determines the final flatness, aims to remove material and planarize the upper and lower surfaces of the silicon wafer through the coordinated movement of upper and lower fixed disks and planetary wheels. However, because the silicon wafer is in a free-floating state without a fixed reference during double-side polishing, existing process monitoring technologies face severe challenges.

[0003] Currently, most mainstream polishing monitoring systems focus on online monitoring of the total thickness of the silicon wafer, typically using non-contact sensors embedded in the mounting plate to acquire the overall change in wafer thickness. While this monitoring method can determine whether the process endpoint has been reached, it cannot analyze the independent material removal status of the upper and lower surfaces. In actual production, due to inconsistencies in the wear states of the upper and lower polishing pads, non-uniform slurry distribution, and slight differences in the mounting plate drive parameters, imbalances in the removal rate on one side are highly likely to occur. Even if the total removal amount meets the set target, this asymmetrical removal behavior will lead to uneven distribution of residual stress inside the silicon wafer, resulting in severe warp or bow deformation, significantly reducing the yield of the finished product.

[0004] While some existing technologies attempt to infer removal rates by monitoring motor current or frictional forces at the contact interface, these methods have significant limitations in the complex environment of chemical mechanical polishing (CMP). Double-sided polishing is a multiphysics coupling process involving solid abrasive cutting and liquid slurry rheology. As the disc rotation speed increases, the hydrodynamic pressure effect between the polishing slurry and the silicon wafer significantly intensifies, producing the so-called "water cushion effect," causing the silicon wafer to tend to migrate from a contact state to a floating state. This hydrodynamic behavior nonlinearly modulates the frictional characteristics of the contact interface, leading to misjudgments by monitoring models that rely solely on acoustic emission or frictional force signals. For example, when excessively high hydrodynamic pressure weakens effective contact, the amplitude of the frictional signal acquired by the sensor decreases, which existing technologies often misinterpret as a linear fluctuation in the removal rate, failing to recognize that its essence is a change in lubrication conditions. The lack of a decoupling mechanism for the complex coupling relationship between hydrodynamic pressure and mechanical friction makes it difficult for existing monitoring systems to accurately identify and independently control the double-sided removal amount under dynamically changing process conditions.

[0005] Therefore, in order to address the shortcomings of existing technologies, a monitoring system and method for the removal amount on the top and bottom surfaces of a 12-inch silicon wafer during double-sided polishing is proposed. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a monitoring system and method for the removal amount of the upper and lower surfaces during double-sided polishing of 12-inch silicon wafers. This solves the problem that existing double-sided polishing technologies cannot accurately distinguish the independent material removal amounts of the upper and lower surfaces due to interference from the hydrodynamic pressure water pad effect, which leads to warping and deformation of the silicon wafers due to asymmetric material removal.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a monitoring system and method for monitoring the removal amount on both sides of a 12-inch silicon wafer during double-sided polishing. First, it synchronously acquires real-time total thickness signals, interfacial mechanical friction signals between the upper and lower mounting pads, and hydrodynamic pressure signals between the upper and lower mounting pads during the double-sided polishing process. The interfacial mechanical friction signal is obtained using a high-frequency acoustic emission sensor closely attached to the back of the polishing pad or embedded in the mounting pad substrate. This signal directly reflects the microscopic cutting and scratching action when the rough peaks of the polishing pad contact the silicon wafer surface. The hydrodynamic pressure signal is obtained using a fluid pressure sensor located at the bottom of the polishing pad's guide grooves or below the permeable layer. This signal reflects the hydrodynamic pressure support effect generated when the polishing fluid undergoes shear flow at the contact interface.

[0008] Secondly, based on the interface mechanical friction signal, the mechanical friction energy indexes of the upper and lower fixed plate sides are extracted respectively. Specifically, the system performs bandpass filtering on the acquired raw signal to remove low-frequency ambient noise and extract the cutting main frequency band signal. Within a preset time sliding window, the root mean square energy of the cutting main frequency band signal is calculated. The mechanical friction energy index of the upper fixed plate side... Mechanical friction energy index with the lower plate side The calculation formula is expressed as follows:

[0009] ;

[0010] In the formula, This represents the number of sampling points within the sliding window. This represents the amplitude of the filtered acoustic emission signal.

[0011] Furthermore, based on the hydrodynamic pressure signal, lubrication state correction coefficients are constructed for the upper and lower fixed plate sides, respectively. The inventors have discovered that as the hydrodynamic pressure increases, the contact interface transitions from boundary lubrication to mixed lubrication or fluid lubrication, leading to an enhanced "water cushion effect" and a reduced effective solid-state contact area. At this point, even if the acoustic emission signal intensity remains constant, the actual material removal efficiency will decrease due to the dispersion of contact stress. Therefore, correction coefficients must be introduced. The lubrication state correction coefficients are then established. With net hydrodynamic pressure increment Nonlinear mapping relationship between them:

[0012] ;

[0013] In the formula, This is the difference between the measured fluid dynamic pressure signal and the equipment's no-load reference fluid pressure. It should be noted that... and These are pre-calibrated process constants. The calibration method involves experimentally determining the Stribeck curve characteristics under different combinations of rotational speed and pressure, and then fitting the result. This formula ensures... Follow The value increases and then decreases monotonically, which is consistent with the principles of tribology.

[0014] Next, the mechanical friction energy index is weighted and corrected using a lubrication condition correction coefficient, and the effective removal vector index for the upper and lower fixed plate sides is calculated. This index... This characterizes the true material cutting power density after eliminating fluid support interference. Effective removal of vector exponent on the upper plate side. Effective removal of vector index from the lower plate side Calculated separately as follows:

[0015] ;

[0016] ;

[0017] Subsequently, based on the proportion of the effective removal vector index on the upper fixed plate side to the effective removal vector index on the lower fixed plate side in the sum of the two, a removal allocation ratio is constructed. Then, combined with the change in the real-time total thickness signal, the removal amounts on the upper and lower surfaces are calculated separately. The removal allocation ratio on the upper fixed plate side is defined. for:

[0018] ;

[0019] In the formula It is a very small positive number. Based on the total thickness change. The independent upper surface removal amount was calculated. With lower surface removal amount :

[0020] ;

[0021] ;

[0022] Finally, the method can further include closed-loop feedback control. When the absolute value of the difference between the amount of material removed from the upper and lower surfaces exceeds a preset asymmetry threshold, the system generates control commands to adjust the rotational speed or back pressure of the upper or lower platen to achieve symmetrical removal.

[0023] A second aspect of the present invention provides a system for monitoring the removal amount on the top and bottom surfaces of a 12-inch silicon wafer during double-sided polishing, for performing the above-described method. The system includes:

[0024] Macro thickness monitoring unit, configured to acquire the total thickness signal of silicon wafer in real time;

[0025] The micro-contact state decoupling sensing unit includes an acoustic emission sensor and a fluid pressure sensor embedded in the upper and lower fixed plates, respectively. The fluid pressure sensor is installed at the bottom of the guide groove of the polishing pad on the fixed plate surface, or below the breathable layer, and transmits pressure through the polishing fluid medium or the breathable structure.

[0026] The data processing controller is configured to receive signals from each sensor and execute the aforementioned data processing and calculation algorithms.

[0027] This invention provides a system and method for monitoring the amount of material removed from the top and bottom surfaces during double-sided polishing of a 12-inch silicon wafer. It offers the following advantages:

[0028] 1. This invention innovatively introduces a fluid dynamic pressure monitoring and lubrication state correction model. By quantifying the fluid dynamic pressure increment generated by the rotation of the fixed plate in real time, a dynamic compensation mechanism is constructed to address the nonlinear characteristics of the Stribeck curve. This mechanism can accurately eliminate friction signal distortion caused by the "water cushion effect" during high-speed polishing, effectively preventing the system from misinterpreting signal attenuation caused by the "floating" of the silicon wafer as a decrease in the removal rate. Thus, even under mixed lubrication conditions, it can accurately capture the true mechanical cutting intensity of the abrasive particles on the silicon wafer surface.

[0029] 2. This invention employs a microscopic contact state decoupling sensing unit to spatially isolate and collect signals from the upper and lower fixed-plate interface. Combined with a weighted allocation algorithm, the macroscopic total thickness change is decomposed into two independent surfaces. This technical solution completely changes the traditional black-box approach of double-sided polishing, which can only monitor the "total removal amount." It endows the equipment with the ability to identify single-sided processing behavior, providing process engineers with absolute removal amount data for each of the upper and lower surfaces, filling the technical gap of lacking layered measurement methods in double-sided processing.

[0030] 3. Based on real-time calculation of asymmetry removal index, this invention constructs a millisecond-level closed-loop feedback control loop that can dynamically adjust the platen rotation speed or back pressure parameters according to the removal difference between the upper and lower surfaces. This proactive intervention mechanism corrects the imbalance of removal rate in real time during processing, rather than discovering the problem only after the process is completed. It suppresses the uneven internal stress caused by over-polishing on one side from the physical source, significantly reducing the warpage and curvature of 12-inch large-size silicon wafers, and achieving dual control over thickness and morphology indicators.

[0031] 4. This invention integrates an adaptive calibration strategy for the consumable lifecycle, which can automatically update the fluid load sensitivity coefficient based on the cumulative usage time of the polishing pad. This design fully considers the long-term impact of polishing pad surface groove wear and micropore blockage on the mapping relationship between the fluid field and the friction field, overcoming the defect of the static model's significant accuracy drift in the later stages of consumable aging, and ensuring that the monitoring system maintains a high level of predictive confidence throughout the entire lifecycle from new pad break-in to old pad retirement.

[0032] 5. The sensing unit of this invention employs a permeation measurement method based on Pascal's principle using a channeling trench and a back-end sensing structure based on acoustic impedance matching, achieving non-invasive "bypass" capture of physical field signals. This arrangement avoids the risk of scratches or edge chipping that may be caused by direct sensor contact with the silicon wafer edge, and utilizes a fixed substrate to physically isolate precision electronic components from highly corrosive chemimechanical slurries, greatly improving the hardware reliability and service life of the system in harsh industrial production environments. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the overall structure of the 12-inch silicon wafer double-sided polishing top and bottom surface removal amount monitoring system of the present invention.

[0034] Figure 2 This is a schematic diagram of the cross-sectional structure of the micro-contact state decoupling sensing unit of the present invention embedded in the plate and polishing pad. Detailed Implementation

[0035] The technical solutions in the embodiments of the present invention will be clearly and completely described below 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.

[0036] Please see the appendix Figure 1 To be continued Figure 2 This invention provides a system and method for monitoring the removal amount on the top and bottom surfaces of a 12-inch silicon wafer during double-sided polishing. The system is configured to work in conjunction with the host of a double-sided polishing equipment.

[0037] The main unit of the double-sided polishing equipment typically includes an upper fixed plate, a lower fixed plate, polishing pads respectively attached to the surfaces of the upper and lower fixed plates, and a planetary wheel located between the upper and lower fixed plates to support the silicon wafer. The monitoring system, in its hardware architecture, mainly consists of a macroscopic thickness monitoring unit, a microscopic contact state decoupling sensing unit, and a data processing controller.

[0038] The macroscopic thickness monitoring unit is configured to acquire real-time global geometrical changes of the silicon wafer during the polishing process. This unit includes a non-contact thickness sensor embedded in the upper or lower mounting plate substrate. The thickness sensor employs a capacitive displacement sensor or an infrared spectral interferometer, and its sensing end face is aligned with the silicon wafer trajectory through a light-transmitting window or unshielded area on the polishing pad, thereby outputting a real-time total thickness signal. .

[0039] The micro-contact state decoupling sensing unit is configured to capture the micro-physical field characteristics of the polished interface, and it further includes an interface mechanical friction monitoring subunit and a hydrodynamic pressure monitoring subunit.

[0040] The interface mechanical friction monitoring subunit includes acoustic emission sensors respectively disposed on the upper and lower fixed plate sides. The acoustic emission sensors use piezoelectric ceramic (PZT) or polyvinylidene fluoride (PVDF) films with high-frequency response characteristics as sensing elements. To ensure high-fidelity signal transmission, the acoustic emission sensors are rigidly bonded to the surface of the fixed plate metal substrate and positioned adjacent to the attachment surface of the polishing pad, or directly encapsulated within the backing layer of the polishing pad. This arrangement allows the sensors to sensitively capture the high-frequency stress waves generated when abrasive particles cut the atomic layers of the silicon wafer surface, i.e., the interface mechanical friction signal.

[0041] The fluid dynamic pressure monitoring subunit includes fluid pressure sensors respectively disposed on the upper and lower fixed plate sides. These fluid pressure sensors are either miniature MEMS pressure sensors or fiber Bragg grating pressure sensors. Their mounting positions are located on the fixed plate surface, corresponding to the bottom of the guide trench of the polishing pad, or below the breathable layer of the polishing pad. This mounting method ensures that the sensor's sensing surface does not directly contact the silicon wafer mechanically, but rather senses fluid pressure through the polishing slurry medium filled within the trench. This sensor is used to output in real-time a fluid dynamic pressure signal characterizing the fluid dynamic pressure generated in the gap between the fixed plate and the silicon wafer due to the relative movement of the polishing slurry.

[0042] The data processing controller is connected to the aforementioned sensing units via a high-speed fieldbus or analog signal acquisition card to achieve synchronous acquisition of multi-channel signals, time-domain alignment, and multi-physics coupling calculation.

[0043] The method for monitoring the removal amount on both sides of a 12-inch silicon wafer during double-sided polishing mainly includes steps such as signal acquisition and preprocessing, feature index extraction, lubrication state correction, and removal amount decoupling.

[0044] First, the synchronous acquisition and preprocessing steps of multiphysics field signals are performed. After the polishing process starts, the data processing controller operates at a preset sampling frequency. Synchronously record the real-time total thickness signal from the macro thickness monitoring unit. Raw acoustic emission signal from the upper fixed plate of the interface mechanical friction monitoring subunit The original signal emitted by the lower fixed plate side acoustic emission And the fluid dynamic pressure signal from the upper plate side of the fluid dynamic pressure monitoring subunit. With the lower plate side fluid dynamic pressure signal Because the original acoustic emission signal is mixed with low-frequency environmental noise generated by the rotation of the equipment motor, bearing friction, and coolant flow, the data processing controller has a built-in digital bandpass filter. This filter is configured to allow only signals in a specific frequency band to pass through, which corresponds to the characteristic frequency range when the abrasive grains and the monocrystalline silicon material undergo material removal, thereby obtaining a pure cutting main frequency band signal.

[0045] Secondly, the mechanical friction energy index is extracted. This step aims to convert the high-frequency waveform signal into a quantitative indicator characterizing the intensity of contact work. The data processing controller uses a sliding time window algorithm to process the cutting main frequency band signal. The sliding window length is set to include... The sampling point, for the upper fixed plate side, at the sampling point... At each discrete sampling time, the root mean square energy of the signal within the window is calculated as the mechanical friction energy index. The calculation formula is as follows:

[0046]

[0047] Similarly, the mechanical friction energy index on the lower plate side The calculation formula is as follows:

[0048]

[0049] In the formula, and These represent the discrete amplitudes of the filtered upper and lower fixed-plate side acoustic emission signals at corresponding times. This index... Physically, it reflects the effective collision and scratch intensity of solid-state contact at the interface per unit time. Under ideal dry friction or boundary lubrication conditions, this index is linearly positively correlated with the material removal rate; however, in actual double-sided polishing conditions, due to the presence of the slurry fluid film, this linear relationship is modulated by hydrodynamic pressure, thus requiring a subsequent hydrodynamic pressure correction step.

[0050] Next, the calibration and net increment calculation steps for the fluid dynamic pressure reference value are performed. The fluid pressure measured during polishing includes the centrifugal back pressure caused by the high-speed rotation of the platen and the static pressure head of the piping system. To obtain the dynamic pressure effect generated solely by shear flow, the background value needs to be subtracted. Under no-load conditions, the data processing controller records the steady-state readings of the fluid pressure sensors on the upper and lower platens, which are then recorded as the reference fluid pressure. and During polishing, calculate the net hydrodynamic pressure increment at the current moment. and :

[0051]

[0052]

[0053] in, and This represents the measured fluid dynamic pressure signal value at the current moment. Net fluid dynamic pressure increment. This value directly reflects the load-bearing capacity of the fluid film between the silicon wafer and the polishing pad. The larger the value, the thicker the fluid film or the stronger the fluid support. In this case, the silicon wafer tends to "float" and the effective penetration depth of the solid abrasive particles into the silicon wafer becomes shallower, resulting in a decrease in mechanical removal efficiency.

[0054] In this embodiment, after obtaining the net hydrodynamic pressure increments on the upper and lower fixed disk sides, a step of constructing the lubrication state correction coefficient is further performed. The core logic of this step lies in identifying and quantifying the inhibitory effect of hydrodynamic action on solid mechanical removal. During chemical mechanical polishing, as the hydrodynamic pressure increases, the contact interface gradually migrates from a boundary lubrication state to a mixed lubrication state or even a fluid lubrication state. This state migration leads to the so-called "water cushion effect," which means that although there is still relative motion between the fixed disk and the silicon wafer, the effective abrasive cutting depth becomes shallower, and the actual material removal efficiency decreases.

[0055] Therefore, this invention establishes a lubrication state correction coefficient. With net hydrodynamic pressure increment A nonlinear mapping model between them. This model is constructed based on the tribological Stribeck curve characteristics, where the correction coefficients... It is defined as a monotonically decreasing function of the net hydrodynamic pressure increment. Specifically, it is the lubrication condition correction factor on the upper platen side. Correction factor for lubrication condition on the lower plate side Calculate according to the following formula:

[0056]

[0057]

[0058] In the formula, Defined as the fluid load sensitivity coefficient, this coefficient is related to the compressive modulus of the polishing pad and the drainage capacity of the surface grooves, and is used to characterize the influence weight of fluid pressure on the contact state under a specific consumable combination; Defined as a nonlinear exponent, it is used to fit the nonlinear decay relationship between fluid dynamic pressure and effective contact area. These two coefficients, as process constants, are predetermined through calibration experiments on standard samples and stored in the data processing controller before the polishing process begins.

[0059] Next, the effective removal vector index is calculated. While the original extracted mechanical friction energy index reflects the vibration intensity of the interface, it does not eliminate the interference from fluid damping and ineffective friction. By introducing a lubrication state correction coefficient, the system corrects the simple mechanical vibration energy to an effective power density that truly characterizes the material removal capability. The effective removal vector index on the upper plate side... Effective removal vector index of the lower plate side The calculations are as follows:

[0060]

[0061]

[0062] This step realizes the physical quantity conversion from "interfacial friction characteristics" to "effective cutting work", ensuring that the evaluation index can still accurately reflect the actual removal trend even under conditions of slurry supply fluctuations or uneven fluid distribution.

[0063] Subsequently, the present invention performs the decoupling calculation steps of constructing the removal allocation ratio and the single-sided removal amount. Since the total removal amount of the upper and lower surfaces during the double-sided polishing process is determined by the total thickness change measured by the macro thickness monitoring unit, this embodiment uses the relative proportion of the effective removal vector index of the upper and lower surfaces to spatially allocate the total removal amount.

[0064] First, calculate the removal allocation ratio on the upper fixing side. :

[0065]

[0066] In the formula, This is a preset, extremely small positive number to prevent calculation singularities caused by the denominator being zero at the moment of equipment startup or under extremely low friction conditions. Correspondingly, the removal allocation ratio on the lower stationary side... Then it is .

[0067] Furthermore, this is combined with the total thickness change output in real time by the macroscopic thickness monitoring unit. The system outputs the independent absolute removal amounts of the upper and lower surfaces through an inversion algorithm. and :

[0068]

[0069]

[0070] Through the above calculation process, this system successfully resolved the previously invisible single-sided removal data in a double-sided polishing environment lacking a fixed geometric reference surface.

[0071] Furthermore, the present invention includes a closed-loop feedback control step based on asymmetric deviation. The data processing controller calculates the amount of material removed from the upper surface in real time. With lower surface removal amount The absolute value of the difference between the two surfaces is used. When this absolute value exceeds a preset asymmetry threshold, the system determines that the current process has a one-sided removal imbalance. At this time, the controller generates a corresponding compensation command based on the sign of the difference. If the upper surface is removed too quickly, the controller will output a command to fine-tune and reduce the rotation speed of the upper mounting plate or reduce the back pressure of the upper airbag carrier; conversely, if the lower surface is removed too quickly, the parameters of the lower mounting plate will be adjusted accordingly. This dynamic closed-loop adjustment mechanism continues to operate until the removal rates of the upper and lower surfaces return to the equilibrium range, thereby effectively preventing warping deformation of the silicon wafer caused by asymmetric removal.

[0072] In this embodiment, to ensure the calculation accuracy of the aforementioned lubrication state correction model, the fluid load sensitivity coefficient is involved. With nonlinear exponent The process constants need to be determined in advance through a standardized calibration process. Since different types of polishing pads have different surface groove textures and compressive moduli, and different slurry compositions have different viscosity characteristics, these factors directly determine the degree of interference of hydrodynamic pressure on solid-state contact. Therefore, this invention employs an offline fitting method based on Stribeck curve characteristics to obtain the aforementioned process constants.

[0073] The specific calibration process is as follows: Before formal production, a full-factor experiment was conducted using the same consumable combination as the production process. By gradually changing the disc rotation speed and loading pressure, various hydrodynamic pressure conditions covering the boundary lubrication to mixed lubrication range were artificially created. Under each condition, the stable net hydrodynamic pressure increment was recorded. The actual material removal rate was measured. The normalized material removal rate data was then fitted with the net hydrodynamic pressure increment data using a nonlinear least squares method to solve for the value that minimizes the error. and Preferred value. This calibration value is then permanently stored in the parameter database of the data processing controller as a baseline parameter for subsequent online monitoring.

[0074] Furthermore, this invention fully considers the wear characteristics of polishing pads during long-term use. As the cumulative usage time of the polishing pad increases, the microporous structure on its surface gradually becomes blocked, and the depth of the drainage grooves becomes shallower due to wear. This gradual change in physical morphology leads to a decrease in the drainage capacity of the polishing pad, thereby altering the mapping relationship between fluid dynamic pressure and contact state. If the initially calibrated parameters are always used, the calculation error of the monitoring model will gradually increase as the consumable's lifespan declines.

[0075] To this end, this system further integrates a parameter adaptive calibration mechanism. The data processing controller records the cumulative usage time of the polishing pad in real time. The system dynamically adjusts the fluid load sensitivity coefficient based on a preset wear drift model. The corrected sensitivity coefficient... The drift compensation formula is used for periodic updates:

[0076]

[0077] In the formula, These are the initial calibration values ​​under the new pad condition; The wear drift coefficient is obtained by analyzing the correlation between the wear rate of the polishing pad and the decay rate of its fluid carrying capacity. By introducing a compensation term in this time dimension, the system can automatically adapt to the full life cycle characteristics of the polishing pad from the "break-in period" to the "stable period" and then to the "aging period," ensuring that the removal quantity decoupling algorithm always maintains high confidence in the dynamic production environment where the condition of consumables is constantly changing.

[0078] In summary, this invention constructs a robust monitoring system with time-varying adaptability by combining offline precision calibration and online adaptive compensation. This system not only corrects the nonlinearity of transient friction signals using fluid pressure during a single polishing process, but also corrects systematic deviations caused by consumable aging over long production cycles. This multi-scale error elimination mechanism enables double-sided polishing equipment to accurately identify the independent removal behavior of the upper and lower surfaces in complex interference environments, providing core metrological support for achieving ultra-precision machining with nanometer-level flatness.

[0079] In this embodiment, based on the calculated removal amounts of the upper and lower surfaces, the data processing controller further executes closed-loop feedback control and polishing endpoint determination steps for asymmetric removal. This step aims to convert the monitored removal amount difference into compensation actions of the device drive unit in real time, thereby actively intervening in the polishing process to ensure that when the silicon wafer reaches the target thickness, not only does the total thickness meet the specifications, but its geometry remains within a controlled range.

[0080] In closed-loop feedback control logic, the system first defines the asymmetry removal index at the current moment. This indicator quantifies the imbalance between the material removal rates of the upper and lower surfaces, and its calculation is based on the difference in the cumulative removal amount of the upper and lower surfaces within the current sampling period.

[0081]

[0082] The system sets an asymmetry threshold. This threshold is set according to the flatness specifications of the silicon wafer product. When At that time, the data processing controller activates the proportional-integral control algorithm to generate correction commands for the speed of the upper or lower fixed plate. Taking the adjustment of the upper fixed plate speed as an example, let the first... The basic setting speed of the time plate is The corrected command speed Generated according to the following control law:

[0083]

[0084] In the formula, This is the proportional gain coefficient, used to quickly remove bias in a rapid response. This is the integral gain coefficient, used to eliminate long-term accumulated steady-state errors; the negative sign in parentheses indicates a negative feedback control mechanism: when When the rotation speed is positive, the system automatically reduces the rotational speed of the upper platen, thereby reducing the relative friction speed and mechanical work on the upper surface and suppressing its tendency to be removed too quickly. Conversely, if... If the speed is negative, the system will increase the speed of the upper platen or decrease the speed of the lower platen to restore balance.

[0085] Simultaneously, the monitoring system implements an endpoint determination strategy based on absolute removal volume. Traditional polishing stops only when the total thickness reaches a threshold, ignoring the removal balance in the final stage. In this embodiment, the system sets a target removal volume. As the polishing process nears its end, the controller enters finishing mode and locks the current allocation ratio. And predict the remaining process time based on the real-time removal rate. :

[0086]

[0087] In the formula, This represents the instantaneous rate of change of the total thickness at the current moment. The system continues to monitor until the cumulative removal amount on both the upper and lower surfaces converges within the target tolerance zone. Subsequently, a stop command is sent to the main equipment to complete the polishing process.

[0088] Through the above-described embodiments, this invention overcomes the "blind polishing" problem caused by consumable state drift or uneven fluid distribution in existing technologies. The system not only visualizes the amount removed from one side but also transforms it into an executable control variable, constructing a complete technical closed loop from "multi-physics sensing" to "intelligent decision-making" and then to "precise execution." Experimental data shows that after adopting this control strategy, the total thickness deviation and warpage of the finished silicon wafer exhibit significant decoupling characteristics, effectively improving the processing yield and surface geometric quality of 12-inch large-size silicon wafers.

[0089] In this embodiment, the physical integration and signal transmission mechanism of the micro-contact state decoupling sensing unit in the double-sided polishing equipment are specifically configured to ensure the stability and reliability of signal acquisition under high-speed rotation and chemical corrosion environments. Since the upper and lower fixed plates are in a continuous independent rotation state during the process, and the polishing environment is filled with corrosive chemical mechanical abrasive slurry, the installation and communication method of the sensor are the basic hardware guarantee for realizing the aforementioned monitoring method.

[0090] For the fluid dynamic pressure monitoring subunit, the fluid pressure sensor is specifically designed as a non-invasive embedded structure. Specifically, the fluid pressure sensor is not directly exposed to the working surface of the polishing pad, but is mounted within a pre-machined groove on the surface of the mounting plate's metal substrate, with the groove precisely corresponding to the bottom of the flow channel on the polishing pad surface. The sensor sensing surface and the flow channel are filled with a corrosion-resistant pressure-permeable medium or coupled through the permeable layer structure of the polishing pad itself. This arrangement utilizes Pascal's principle, allowing the polishing fluid filled in the flow channel to transmit fluid dynamic pressure to the sensor sensing surface without damage, while avoiding physical damage caused by direct impact of the silicon wafer edge onto the sensor, thus meeting the process requirements for "bypass monitoring" of the fluid field.

[0091] For the interface mechanical friction monitoring subunit, the acoustic emission sensor is fixed to the area near the edge of the back of the fixed plate using an insulating potting process. To eliminate electromagnetic interference and background vibration generated by the rotation of the fixed plate, an acoustic impedance matching layer is set between the sensor and the fixed plate substrate, and the signal transmission cable adopts a double-shielded structure. Considering the rotation characteristics of the fixed plate, the sensor signal transmission between the upper and lower fixed plate sides is realized through an industrial-grade conductive slip ring or a wireless radio frequency transmission module. The conductive slip ring is installed at the central rotation axis of the fixed plate, realizing the continuous transmission of the sensor electrical signal in the rotating coordinate system to the data processing controller in the stationary coordinate system, ensuring zero packet loss of data across the entire rotational speed range.

[0092] The data processing controller is architecturally divided into a signal synchronization acquisition module, an algorithm calculation core module, and an execution control interface module. The signal synchronization acquisition module is equipped with a high-precision clock source to uniformly timestamp the analog voltage signal from the macroscopic thickness monitoring unit, the charge signal from the acoustic emission sensor, and the digital signal from the fluid pressure sensor, eliminating phase errors caused by response delays from different sensors. The algorithm calculation core module is the system's computational hub, internally storing the aforementioned lubrication state correction model and adaptive calibration parameters. This module executes embedded code to convert the input raw physical signal into the amount of material removed from the upper surface in real time. With lower surface removal amount .

[0093] Furthermore, the monitoring system integrates an abnormal state self-check function. When the reading of the fluid pressure sensor is lower than the preset minimum fluid film-forming pressure threshold for an extended period, or when the signal-to-noise ratio of the acoustic emission signal is lower than the effective detection limit, the data processing controller will determine that the sensor coupling has failed or the polishing pad grooves are severely blocked. At this time, the system triggers an alarm signal and automatically suspends the closed-loop control logic to maintain the currently set speed and pressure parameters, preventing incorrect removal compensation due to input signal distortion, thereby ensuring the safety of the process and the robustness of the system. This hardware and software co-design allows the monitoring system provided by this invention to be seamlessly integrated into existing mass-production 12-inch double-sided polishing machines without requiring destructive modifications to the main structure of the equipment.

[0094] This embodiment details the complete process flow for double-sided polishing of 12-inch silicon wafers using the aforementioned monitoring system, demonstrating the operational timing and control logic of this technical solution in actual industrial production. The process is designed to ensure the total thickness of the silicon wafer meets the standard while strictly controlling wafer warpage caused by uneven removal on one side.

[0095] The process flow begins with the initialization and parameter retrieval phase. Once a batch of silicon wafers is loaded onto the planetary wheel and pressurized by the closing of the upper stationary plate, the data processing controller first reads the cumulative lifespan data of the polishing pads via the communication interface. Then, based on the aforementioned adaptive calibration mechanism, it calls the corrected fluid load sensitivity coefficient for the current state. With nonlinear exponent The system simultaneously performs zero-point drift calibration on both the acoustic emission sensor and the fluid pressure sensor, ensuring the consistency of all monitoring data.

[0096] Subsequently, the coarse polishing monitoring stage begins. The main unit of the equipment starts the platen rotation and slurry supply according to the preset formula. At this time, the macroscopic thickness monitoring unit begins to output the total thickness decrease curve at a high-frequency sampling rate. Simultaneously, the microscopic contact state decoupling sensing unit synchronously captures interface signals. The data processing controller runs the decoupling algorithm in real time, decomposing the simple thickness change into upper surface removal components and lower surface removal components. In this stage, the system mainly focuses on the stability of the removal rate, through real-time calculation of the effective removal vector index. Monitor for any localized dry friction anomalies caused by insufficient slurry distribution.

[0097] When the total removal amount reaches the preset percentage of the overall process target, the system automatically switches to the fine polishing and asymmetric control stage. In this stage, the asymmetric deviation closed-loop feedback control logic has the highest priority. The system calculates the difference in removal amounts between the upper and lower surfaces with an extremely short control cycle. If monitoring data indicates that the removal amount on the upper surface significantly lags behind that on the lower surface, the controller does not change the total pressure. Instead, it adjusts the drive speed of the upper platen or regulates the pressure distribution in the upper back pressure chamber to directionally increase the mechanical friction work density on the upper surface. This dynamic adjustment continues until the real-time calculated asymmetry index is reached. Converging to the allowable tolerance zone Within the range.

[0098] The determination of the process endpoint no longer relies solely on the total thickness signal, but instead employs a dual convergence criterion. The system continuously predicts the remaining processing time and issues a stop command if and only if the following two conditions are met: first, the real-time measured total thickness of the silicon wafer... Achieve target thickness Secondly, the calculated ratio of the cumulative removal amounts on the upper and lower surfaces. The following symmetry constraints must be satisfied:

[0099]

[0100] In the formula, , This is a strict symmetry tolerance related to silicon wafer warpage specifications. This decision logic ensures that at the moment of unloading, the silicon wafer not only meets the thickness requirements but also that its internal residual stress distribution is symmetrical, thereby physically eliminating the cause of "bowl-shaped" or "umbrella-shaped" warpage after polishing.

[0101] Finally, during the cleaning and unloading interval after polishing, the system performs data archiving and model iteration steps. The actual processing data for this batch of silicon wafers, including the average level of hydrodynamic pressure, the attenuation characteristics of the acoustic emission signal, and the final removal rate distribution, are entered into the historical database. The system uses this data to update the wear drift coefficient of the polishing pad. This provides more accurate initial parameters for the processing of the next batch of silicon wafers.

[0102] In summary, the technical solution provided by this invention successfully establishes a pair of "eyes" capable of seeing the independent behavior of the upper and lower surfaces in the "black box" double-sided processing environment by introducing a multi-physics field decoupling mechanism based on hydrodynamic pressure correction into the traditional double-sided polishing process. This not only solves the long-standing problem of removing asymmetry in large-size silicon wafer processing, but also significantly improves the equipment's tolerance to changes in consumable conditions and process stability through closed-loop control and adaptive calibration throughout the entire process, demonstrating extremely high industrial application value.

[0103] 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 monitoring the removal amount on the top and bottom surfaces of a 12-inch silicon wafer during double-sided polishing, characterized in that, Includes the following steps: Simultaneously acquire real-time total thickness signal, interface mechanical friction signal between the upper and lower fixed plate sides, and hydrodynamic pressure signal between the upper and lower fixed plate sides of the silicon wafer during the double-sided polishing process. Based on the interface mechanical friction signal, the mechanical friction energy index of the upper plate side and the lower plate side is extracted respectively. Based on the fluid dynamic pressure signal, lubrication state correction coefficients are constructed for the upper platen side and the lower platen side, respectively. The lubrication state correction coefficients are used to characterize the degree of inhibition of material removal efficiency by fluid support. The mechanical friction energy index is weighted and corrected using the lubrication state correction coefficient, and the effective removal vector index of the upper and lower plate sides is calculated. Based on the ratio of the effective removal vector index on the upper plate side to the effective removal vector index on the lower plate side in the sum of the two, a removal allocation ratio is constructed, and combined with the change in the real-time total thickness signal, the removal amount on the upper surface and the removal amount on the lower surface are calculated respectively.

2. The method according to claim 1, characterized in that, The interface mechanical friction signal is a high-frequency acoustic emission signal; The step of extracting the mechanical friction energy index of the upper and lower fixed plate sides respectively specifically includes: The interface mechanical friction signals between the upper and lower fixed plates are subjected to frequency band filtering to extract the cutting main frequency band signal. Within a preset time sliding window, the root mean square energy of the cutting main frequency band signal is calculated respectively, and the calculated root mean square energy is used as the mechanical friction energy index.

3. The method according to claim 1, characterized in that, The step of constructing the lubrication state correction coefficients for the upper and lower fixed plate sides respectively specifically includes: Obtain the reference fluid pressure when the equipment is running under no-load, calculate the difference between the fluid dynamic pressure signal collected at the current moment and the reference fluid pressure, and obtain the net fluid dynamic pressure increment. A nonlinear mapping relationship is established between the lubrication state correction coefficient and the net hydrodynamic pressure increment, wherein the lubrication state correction coefficient decreases monotonically as the net hydrodynamic pressure increment increases.

4. The method according to claim 3, characterized in that, The establishment of the nonlinear mapping relationship between the lubrication state correction coefficient and the net fluid dynamic pressure increment is specifically manifested as follows: The lubrication condition correction coefficient is equal to 1 divided by the denominator, where the denominator is 1 plus the product of the fluid load sensitivity coefficient and the nonlinear exponent of the net fluid dynamic pressure increment. Wherein, the fluid load sensitivity coefficient and the nonlinear index are preset process constants.

5. The method according to claim 1, characterized in that, The step of using the lubrication state correction coefficient to weight and correct the mechanical friction energy index, and calculating the effective removal vector index for the upper and lower fixed plate sides, specifically includes: Multiply the mechanical friction energy index of the upper plate side by the lubrication state correction coefficient of the upper plate side to obtain the effective removal vector index of the upper plate side; Multiply the mechanical friction energy index of the lower platen side by the lubrication state correction coefficient of the lower platen side to obtain the effective removal vector index of the lower platen side.

6. The method according to claim 5, characterized in that, The step of constructing the removal allocation ratio based on the ratio of the effective removal vector index on the upper positioning side to the effective removal vector index on the lower positioning side in the sum of the two specifically includes: Divide the effective removal vector index of the upper fixed plate side by the sum of the effective removal vector indices of the upper fixed plate side and the lower fixed plate side to obtain the removal allocation ratio of the upper fixed plate side. The removal allocation ratio of the lower platen side is obtained by subtracting the removal allocation ratio of the upper platen side from 1.

7. The method according to claim 1, characterized in that, It also includes a closed-loop feedback control step: calculating the absolute value of the difference between the amount removed from the upper surface and the amount removed from the lower surface; When the absolute value of the difference exceeds the preset asymmetry threshold, a speed adjustment command or a pressure adjustment command is generated according to the positive or negative sign of the difference, and the driving parameters of the upper plate or the lower plate are independently compensated until the removal rate of the upper surface and the removal rate of the lower surface tend to be balanced.

8. The method according to claim 3, characterized in that, The fluid dynamic pressure signal is acquired from a fluid pressure sensor installed at the bottom of the guide groove or below the breathable layer of the polishing pad on the surface of the plate. The sensing surface of the fluid pressure sensor does not directly contact the silicon wafer, but transmits pressure through the polishing fluid medium.

9. The method according to claim 4, characterized in that, It also includes a parameter adaptive calibration step: Record the cumulative usage time of the polishing pad or the number of silicon wafers processed; Based on the cumulative usage time or the number of silicon wafers processed, the fluid carrying sensitivity coefficient in the nonlinear mapping relationship is dynamically adjusted to compensate for the drift of hydrodynamic characteristics caused by polishing pad wear.

10. A system for monitoring the removal amount on both the top and bottom surfaces of a 12-inch silicon wafer during double-sided polishing, characterized in that, include: Macro thickness monitoring unit, configured to acquire the total thickness signal of silicon wafer in real time; The micro-contact state decoupling sensing unit includes an acoustic emission sensor and a fluid pressure sensor embedded in the upper and lower fixed plates, respectively, and is configured to synchronously acquire the interface mechanical friction signal and fluid dynamic pressure signal between the upper and lower fixed plates. A data processing controller is configured to receive signals from the macroscopic thickness monitoring unit and the microscopic contact state decoupling sensing unit, and execute the method as described in any one of claims 1 to 9 to output independent upper surface removal amount and lower surface removal amount.

Images