A low-noise preamplification type semiconductor low-frequency noise test system

Abstract

The application relates to the technical field of semiconductor device testing, and discloses a low-noise pre-amplification type semiconductor low-frequency noise testing system, which comprises a precise biasing and comb excitation unit, an adaptive analog equalization pre-amplification unit, a data acquisition unit and a main control and spectrum separation and de-embedding unit; the precise biasing unit sets a static working point in response to an instruction and injects a frequency domain sparse comb excitation superimposed on direct current; the adaptive amplification unit uses an analog filter network to perform spectrum shaping on a mixed signal containing intrinsic noise; and the main control unit calculates a system transfer function in real time based on the response amplitude of a pilot frequency point and performs de-embedding processing on noise data of non-pilot points through interpolation fitting. Through the frequency domain sparse pilot excitation and analog domain pre-whitening technology, the low-noise pre-amplification type semiconductor low-frequency noise testing system realizes the synchronization of transfer function calibration and device noise acquisition in single measurement, and effectively solves the problems of gain drift and limited 1 / f noise dynamic range in wideband testing.

Description

Technical Field

[0001] This invention relates to the field of semiconductor device testing technology, specifically a low-noise preamplifier semiconductor low-frequency noise testing system. Background Technology

[0002] Low-frequency noise characteristics of semiconductor devices are crucial indicators for evaluating device fabrication quality, reliability, and surface state density. Accurately obtaining the noise power spectral density at the device's input during low-frequency noise testing depends not only on acquiring the output noise signal but also on the precise calibration of the overall transfer function of the test system. Existing testing methods typically separate noise measurement from system gain calibration in time. This involves first calibrating the gain and frequency response of the amplification link using a known signal source, then disconnecting the signal source and connecting it to the device under test for noise measurement, or using a single-point frequency successive scan method to obtain impedance matching characteristics. This step-by-step operation not only leads to low testing efficiency, failing to meet the needs of large-scale wafer-level testing, but more importantly, amplifier gain and bandwidth often drift with temperature, and the device's output impedance changes with bias conditions. Step-by-step calibration cannot reflect the true system load effect and gain state at the measurement moment, resulting in errors in the final calculated input noise data.

[0003] On the other hand, low-frequency noise in semiconductor devices typically exhibits a 1 / f characteristic, meaning the noise amplitude at low frequencies is much higher than at high frequencies. This spectral distribution, which increases with decreasing frequency, poses a significant challenge to the dynamic range of signal conditioning circuits. Existing broadband preamplifiers typically employ a fixed-gain mode across the entire frequency band. If a high gain is set to meet the acquisition requirements of weak thermal noise in the high-frequency band, the high-amplitude flicker noise in the low-frequency band will cause the final stage of the amplifier or the analog-to-digital converter to enter a saturated state. Conversely, if the system gain is reduced to prevent low-frequency saturation, the signal amplitude in the high-frequency band often falls below the quantization noise floor of the analog-to-digital converter, making it impossible to effectively extract high-frequency noise. This fixed-gain architecture struggles to accommodate the amplitude differences of 1 / f noise signals across a wide frequency band, limiting the effective measurement bandwidth and resolution of the test system.

[0004] Furthermore, since low-frequency noise signals in semiconductors are typically in the nanovolt to microvolt range, they are susceptible to interference from external environmental factors and system non-ideal factors. During long-term noise waveform acquisition, if the clock of the excitation signal source and the clock of the data acquisition system are not strictly synchronized, non-integer cycle sampling truncation will cause spectral leakage during frequency domain analysis, resulting in the side lobes of the excitation signal masking weak noise signals at adjacent frequencies. Simultaneously, when processing weak current signals, parasitic leakage current paths on the surface of the printed circuit board's insulating substrate introduce measurement errors. Conventional passive shielding measures are insufficient to completely block this leakage interference in parallel with the signal path, further reducing the confidence level of the test data. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a low-noise preamplifier semiconductor low-frequency noise testing system. This system solves the problems in existing semiconductor low-frequency noise testing technologies, such as limited efficiency of traditional point-frequency testing, difficulty in real-time calibration of the system transfer function in full-band testing, and the tendency of broadband preamplifiers to saturate or have insufficient quantization accuracy when processing high dynamic range 1 / f noise.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a low-noise preamplifier semiconductor low-frequency noise testing system, comprising a precision bias and comb excitation unit, an adaptive analog equalization preamplifier unit, a data acquisition unit, and a main control and spectrum separation / de-embedding unit. The analog output terminal of the precision bias and comb excitation unit is connected to the input electrode of the semiconductor device, outputs a DC bias voltage in response to a bias command, determines the static operating point of the semiconductor device, and superimposes a frequency-domain sparse comb excitation signal generated based on waveform synthesis data onto the DC bias voltage and injects it into the semiconductor device. The signal input terminal of the adaptive analog equalization preamplifier unit is connected to the output electrode of the semiconductor device, receives a mixed output signal containing the device's intrinsic noise and the response to the comb excitation signal, and performs analog domain spectrum shaping on the mixed output signal through an internal analog filter network with adjustable frequency response characteristics. The data acquisition unit is connected to the adaptive analog equalization preamplifier unit, which converts the shaped mixed output signal into a digital time-domain sequence. The main control and spectrum separation / de-embedding unit is connected to each of the above units, sends bias commands and waveform synthesis data to the precision bias and comb excitation unit, and receives the digital time-domain sequence transmitted by the data acquisition unit; the main control and spectrum separation / de-embedding unit calculates the system transfer function based on the response amplitude of the comb excitation signal at the pilot frequency point, and performs de-embedding processing on the noise data at non-pilot frequency points to calculate the input noise power spectrum of the semiconductor device.

[0007] Preferably, the precision bias and comb-shaped excitation unit includes, in its physical circuit architecture, a digital isolation interface, a dynamic excitation generation module, a precision attenuation network, and a signal superposition driving circuit connected in sequence, and is independently equipped with a DC reference voltage source. The dynamic excitation generation module converts the waveform synthesis data into an analog stepped wave, and outputs the original AC excitation signal after filtering. The precision attenuation network attenuates the amplitude of the original AC excitation signal by a fixed ratio. The signal superposition driving circuit adopts a fully DC-coupled topology, and its input terminals are respectively connected to the DC reference voltage source and the precision attenuation network, linearly synthesizing the DC bias voltage and the attenuated AC excitation signal. This architecture utilizes a high-precision DAC to generate a small-amplitude AC excitation and superimposes it onto a low-noise DC bias, avoiding the low-frequency cutoff phenomenon introduced by traditional AC coupling capacitors.

[0008] Preferably, the main control and spectrum separation / de-embedding unit divides the frequencies within the measurement bandwidth into a pilot frequency set and a noise observation frequency set. When generating the comb-shaped excitation signal, the main control and spectrum separation / de-embedding unit sets the amplitude of the frequency points belonging to the pilot frequency set to a non-zero value, where the non-zero value is selected as the smaller of the amplitude value calculated based on the signal-to-noise ratio requirement and the amplitude average value calculated based on the total power limit. The amplitude of the frequency points belonging to the noise observation frequency set is forcibly set to zero. This scheme constructs discrete pilot points and continuous noise observation intervals in the frequency domain, enabling the excitation signal to stimulate the system response for calibration, while avoiding interference with the measurement data of the noise observation frequency points.

[0009] Preferably, the adaptive analog equalization preamplifier unit adopts a cascaded topology, including a first-stage fixed-gain low-noise amplifier stage, an inter-stage AC coupling network, and a second-stage programmable shaping amplifier stage. The first-stage fixed-gain low-noise amplifier stage includes input transistor pairs, which are composed of multiple pairs of parallel low-noise junction field-effect transistors forming a differential common-source topology. The second-stage programmable shaping amplifier stage is located after the inter-stage AC coupling network, and a programmable impedance network is integrated in the feedback loop of the second-stage programmable shaping amplifier stage. The main control and spectrum separation / de-embedding unit adjusts the programmable impedance network according to the estimated noise power spectral density. The second-stage programmable shaping amplifier stage has a frequency response characteristic opposite to that of the semiconductor device noise spectrum. This design compensates for the 1 / f characteristic of semiconductor device noise increasing with decreasing frequency through analog domain spectrum shaping, making the signal spectrum output to the data acquisition unit more flat and improving the effective dynamic range of the measurement link.

[0010] Preferably, the programmable impedance network consists of a precision resistor array, a capacitor array, and a switch matrix; the precision resistor array and the precision capacitor array are connected in parallel, and the switch matrix includes multiple relays; the main control and spectrum separation / de-embedding unit controls the closing and opening of the switch matrix to change the equivalent parallel total resistance and equivalent parallel total capacitance in the feedback loop, thereby adjusting the pole and zero frequencies of the amplifier stage.

[0011] Preferably, the main control and spectrum separation / de-embedding unit performs interpolation fitting of the system transfer function; specifically, this includes: extracting the discrete system frequency response at the pilot frequency point, performing amplitude-phase separation and phase dewinding processing on the discrete system frequency response; and using a cubic spline interpolation algorithm to perform interpolation operations on the amplitude frequency response and the dewinding phase frequency response respectively, generating a continuous full-band interpolation transfer function covering the entire measurement bandwidth. This step reconstructs the full-band system transmission characteristics using discrete pilot response data, correcting the error of system gain variation with frequency in broadband measurements.

[0012] Preferably, when performing the de-embedding process, the main control and spectrum separation and de-embedding unit performs a spectrum masking operation to remove pilot frequency point data containing comb excitation signal energy and retain noise observation frequency point data; and uses the full-band interpolation transfer function to perform inverse calculation on the noise observation frequency point data to subtract the pre-stored system background noise power spectral density.

[0013] Preferably, the main control and spectrum separation and de-embedding unit performs frequency domain data completion operation; for the pilot frequency points eliminated by the spectrum masking operation, the main control and spectrum separation and de-embedding unit selects the power spectral density values ​​of the noise observation frequency points on the left and right sides adjacent to the pilot frequency point, calculates the weighted average or arithmetic average as the noise power spectral density estimate at the pilot frequency, and obtains a continuous noise power spectrum in the frequency domain.

[0014] Preferably, the precision bias and comb excitation unit includes a digital-to-analog converter (DAC), whose waveform playback clock source is kept in the same source or phase-locked with the sampling clock source of the data acquisition unit. The main control and spectrum separation / de-embedding unit sets the playback period length of the comb excitation signal to be equal to the fast Fourier transform analysis frame length of the data acquisition unit, and applies rectangular window processing to the acquired time-domain data. By synchronizing the system clock and matching the period, the comb spectral line frequencies are aligned with the frequency grid points of the FFT analysis, preventing spectral leakage from interfering with noise measurements.

[0015] Preferably, the signal superposition driving circuit includes an operational amplifier, and an equipotential protection ring is arranged around the output traces of the signal superposition driving circuit; the equipotential protection ring is connected to the follower output terminal of the operational amplifier, and the potential of the equipotential protection ring is consistent with the potential of the output signal line. This structure blocks the leakage current path on the surface of the insulating substrate through active potential following, ensuring the accuracy of small signal transmission.

[0016] This invention provides a low-noise preamplifier semiconductor low-frequency noise testing system. It has the following advantages: 1. This invention utilizes a frequency-domain sparse comb-type pilot excitation technique to achieve simultaneous measurement of semiconductor device noise and calibration of the system transfer function. By superimposing weak AC signals of discrete frequencies onto a DC bias, the main control unit can extract the system's gain and phase response at the pilot points in real time and construct the full-band transfer function using interpolation fitting. This method eliminates calculation errors introduced by amplifier gain temperature drift or device impedance mismatch in traditional testing, eliminates the need for disconnection for independent calibration, and improves the efficiency and accuracy of low-frequency noise measurement.

[0017] 2. This invention employs an adaptive analog equalization preamplifier structure with programmable frequency response characteristics, optimizing the effective dynamic range of the measurement link. Addressing the 1 / f characteristic of semiconductor devices where low-frequency noise amplitude increases with decreasing frequency, the system configures an opposite filter response in the analog domain to pre-whiten the signal. This mechanism prevents high-amplitude low-frequency noise components from saturating the amplifier or analog-to-digital converter, while ensuring sufficient gain and quantization resolution for weak thermal noise components in the high-frequency band.

[0018] 3. This invention ensures the integrity and spectral purity of weak signals through a clock synchronization mechanism and active equipotential protection design. Excitation generation and data acquisition share a clock source and are periodically aligned, ensuring that pilot energy is strictly concentrated on the frequency grid points of the FFT analysis, thus preventing spectral leakage from interfering with nearby noise data. Combined with an equipotential protection ring driven by the op-amp follower output, leakage current paths on the circuit board surface are blocked, ensuring the signal-to-noise ratio of nanovolt to microvolt level noise signal transmission. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the system functional architecture of the present invention; Figure 2 This is a schematic diagram of the system operation process of the present invention; Figure 3 This is a schematic diagram comparing the dynamic range utilization of the ADC of the present invention; Figure 4 This is a schematic diagram showing the consistency comparison of high-frequency noise measurements according to the present invention.

[0020] Among them: 100, precision bias and comb excitation unit; 200, adaptive analog equalization preamplifier unit; 300, data acquisition unit; 400, main control and spectrum separation and de-embedding unit. Detailed Implementation

[0021] 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.

[0022] Please see the appendix Figure 1 This invention provides a low-noise preamplifier semiconductor low-frequency noise testing system for high dynamic range measurement of the low-frequency noise characteristics of semiconductor devices over a wide frequency band.

[0023] This low-noise preamplifier semiconductor low-frequency noise testing system mainly includes: a precision bias and comb excitation unit 100, an adaptive analog equalization preamplifier unit 200, a data acquisition unit 300, and a main control and spectrum separation / de-embedding unit 400. These units work together through electrical connections to form a closed-loop measurement system.

[0024] The control input terminal of the precision bias and comb excitation unit 100 is connected to the main control and spectrum separation / de-embedding unit 400 via a digital control bus, receiving bias setting commands and waveform synthesis data. The analog output terminal of the precision bias and comb excitation unit 100 is connected to the input electrode of the semiconductor device (e.g., the gate of a field-effect transistor or the base of a bipolar transistor) via a low-noise shielded cable. The precision bias and comb excitation unit 100 is used to provide a stable DC static operating point for the semiconductor device and generates a frequency-sparse comb excitation signal according to the main control command. This comb excitation signal is superimposed on the DC bias voltage and injected into the control terminal of the semiconductor device.

[0025] The signal input terminal of the adaptive analog equalization preamplifier unit 200 is connected to the output electrode of a semiconductor device (e.g., the drain of a field-effect transistor or the collector of a bipolar transistor). The configuration interface of the adaptive analog equalization preamplifier unit 200 is connected to the main control and spectrum separation / de-embedding unit 400 via a digital bus. The adaptive analog equalization preamplifier unit 200 receives a mixed output signal from the semiconductor device, which includes the device's intrinsic noise and its response to a comb excitation signal. The adaptive analog equalization preamplifier unit 200 internally includes an analog filter network with adjustable frequency response characteristics for analog-domain spectrum shaping and amplification of the mixed output signal.

[0026] The analog input terminal of the data acquisition unit 300 is connected to the signal output terminal of the adaptive analog equalization preamplifier unit 200 through an impedance matching circuit. The data output interface of the data acquisition unit 300 is connected to the main control and spectrum separation / de-embedding unit 400. The data acquisition unit 300 is used to convert the analog signal processed by the adaptive analog equalization preamplifier unit 200 into a digital time-domain sequence and transmit it to the main control and spectrum separation / de-embedding unit 400.

[0027] The main control and spectrum separation / de-embedding unit 400 is the control core and data processing center of the system. The main control and spectrum separation / de-embedding unit 400 establishes communication connections with the precision bias and comb excitation unit 100, the adaptive analog equalization preamplifier unit 200, and the data acquisition unit 300. The main control and spectrum separation / de-embedding unit 400 is used to perform measurement process control, excitation waveform data generation, filter parameter calculation, system transfer function extraction, and noise power spectrum reconstruction calculation.

[0028] In terms of physical connection, semiconductor devices are typically placed in test fixtures or probe stations with electromagnetic shielding. Both the precision bias and comb excitation unit 100 and the adaptive analog equalization preamplifier unit 200 are powered by low-noise power supplies and share a common ground with the main control and spectrum separation / de-embedding unit 400 to ensure the integrity of the signal loop.

[0029] This low-noise preamplifier semiconductor low-frequency noise testing system forms two main signal loops through the aforementioned connection relationship: one is the excitation loop, consisting of a precision bias and comb excitation unit 100 to the semiconductor device; the other is the measurement loop, consisting of the semiconductor device via an adaptive analog equalization preamplifier unit 200 to the data acquisition unit 300. The main control and spectrum separation / de-embedding unit 400 monitors the output of the measurement loop and adjusts the signal amplitude of the excitation loop and the frequency response parameters of the measurement loop in real time, realizing closed-loop adaptive control of the measurement system.

[0030] See attached document Figure 2 This invention provides a semiconductor low-frequency noise measurement method based on this system. This method achieves adaptive measurement of low-frequency noise in semiconductor devices through the collaboration of hardware and software.

[0031] At the start of the measurement phase, the main control and spectrum separation / de-embedding unit 400 first sends a DC bias command to the precision bias and comb excitation unit 100. The precision bias and comb excitation unit 100 responds to the command by outputting a set DC voltage to the semiconductor device, causing the semiconductor device to enter a preset static operating state. Simultaneously, the main control and spectrum separation / de-embedding unit 400 sends an initialization command to the adaptive analog equalization preamplifier unit 200, configuring the analog filter network inside the adaptive analog equalization preamplifier unit 200 into a flat-gain all-pass mode.

[0032] Subsequently, the system performs a baseline noise detection operation. The data acquisition unit 300 acquires the short-time domain signal output from the adaptive analog equalization preamplifier unit 200 and transmits it to the main control and spectrum separation / de-embedding unit 400. The main control and spectrum separation / de-embedding unit 400 performs a fast Fourier transform on the time-domain signal to calculate a coarse estimate of the semiconductor device's noise power spectral density. The main control and spectrum separation / de-embedding unit 400 identifies the amplitude range and frequency domain attenuation characteristics of this power spectral density estimate.

[0033] Based on the results of the baseline noise detection, the main control and spectrum separation / de-embedding unit 400 calculates and generates frequency-sparse comb-shaped excitation signal data. The main control and spectrum separation / de-embedding unit 400 determines the amplitude of the comb-shaped excitation signal at each pilot frequency point based on the power spectral density estimate. This amplitude setting follows an adaptive principle: at each pilot frequency point, the power of the excitation signal is higher than a preset proportion of the background noise power at the same frequency, while maintaining the total signal power within the small-signal linear response range of the semiconductor device.

[0034] Simultaneously, the main control and spectrum separation / de-embedding unit 400 calculates the target transfer function based on the reciprocal characteristic of the power spectral density estimate. The main control and spectrum separation / de-embedding unit 400 maps the target transfer function to the hardware configuration parameters of the adaptive analog equalization preamplifier unit 200, and sends these parameters to the adaptive analog equalization preamplifier unit 200 via a digital bus. The adaptive analog equalization preamplifier unit 200 adjusts its internal impedance network according to these parameters, causing its physical frequency response to exhibit low-pass filtering characteristics, with a cutoff frequency slightly higher than the upper limit of the current measurement focus frequency band.

[0035] After completing the above configuration, the system enters the closed-loop measurement phase. The precision bias and comb excitation unit 100 generates an analog voltage waveform based on the received comb excitation signal data and superimposes it onto the DC bias, continuously injecting it into the input terminal of the semiconductor device. The semiconductor device outputs a mixed signal containing intrinsic noise and excitation response.

[0036] The mixed signal enters the adaptive analog equalization preamplifier unit 200. The adaptive analog equalization preamplifier unit 200 uses pre-configured impedance network parameters to perform analog-domain shaping on the mixed signal, providing a gain that matches its noise amplitude, while filtering out high-frequency broadband noise and interference outside the measurement frequency band to prevent signal aliasing. The data acquisition unit 300 performs analog-to-digital conversion on the shaped analog signal to acquire a high dynamic range digital sequence.

[0037] The main control and spectrum separation / de-embedding unit 400 receives the digital sequence and performs frequency domain analysis. The main control and spectrum separation / de-embedding unit 400 extracts the signal amplitude at a preset pilot frequency point, and calculates the discrete system transfer function value using the known injected excitation amplitude and the measured response amplitude. The main control and spectrum separation / de-embedding unit 400 uses an interpolation algorithm to fit a continuous system transfer function curve covering the entire measurement frequency band based on the discrete system transfer function value.

[0038] Finally, the main control and spectrum separation / de-embedding unit 400 performs spectrum separation and reconstruction. The main control and spectrum separation / de-embedding unit 400 removes data from pilot frequency points, retaining data from noise observation frequency points. Using the fitted continuous system transfer function, the main control and spectrum separation / de-embedding unit 400 performs inverse calculations on the noise observation frequency point data, eliminating the frequency response effects introduced by the adaptive analog equalization preamplifier unit 200 and the semiconductor device output impedance, thus restoring the true low-frequency noise power spectrum at the semiconductor device input. For the noise data at the removed pilot frequency points, the main control and spectrum separation / de-embedding unit 400 completes the data through interpolation between adjacent frequency points.

[0039] The precision bias and comb-shaped excitation unit 100 includes a digital isolation interface, a DC reference voltage source, a dynamic excitation generation module, a precision attenuation network, and a signal superposition driving circuit in its physical circuit architecture. All of the above circuits are integrated on a multilayer printed circuit board with an independent electromagnetic shielding cavity.

[0040] The digital isolation interface is located at the signal input front end of the unit and uses a high-speed optical coupler or magnetic isolation chip to cut off the electrical connection between the main control and spectrum separation and de-embedding unit 400 and the precision bias and comb excitation unit 100, so as to prevent digital ground noise from being coupled to the analog ground plane through the common mode path.

[0041] A DC reference voltage source is used to establish a low-noise DC reference potential for the system. This voltage source uses a low-noise bandgap reference chip as its core, with a multi-stage RC low-pass filter cascaded at its output. The cutoff frequency of this filter is set in the millihertz (mHz) range, and polypropylene film capacitors or polystyrene capacitors are used to eliminate dielectric absorption effects and leakage current noise, ensuring that the noise power spectral density of the output DC bias voltage is lower than the background noise floor of the semiconductor device under test throughout the entire measurement bandwidth.

[0042] The dynamic excitation generation module is responsible for generating the frequency domain comb wave signal. This module consists of a high-resolution digital-to-analog converter (DAC) and a reconstruction filter. The DAC receives digital waveform data from the isolation interface and converts it into an analog stepped wave. The reconstruction filter is located at the output of the DAC and is configured as a Butterworth or Bessel response type active low-pass filter to smooth the stepped wave and filter out high-frequency components of quantization noise, outputting the original AC excitation signal.

[0043] A precision attenuation network is connected to the output path of the dynamic excitation generation module to address the issues of excessively high DAC noise floor and excessively large output amplitude. This network employs an L-shaped voltage divider structure composed of low-temperature-coefficient metal foil resistors. The attenuation ratio of this voltage divider structure is set to a constant value (e.g., 1000:1 to 10000:1), compressing the amplitude of the original AC excitation signal to the microvolt level and simultaneously attenuating the thermal noise of the DAC output stage below the system thermal noise limit.

[0044] The signal superposition drive circuit is used to achieve linear synthesis of DC bias and weak AC excitation to drive the device under test (DUT). This circuit employs a fully DC-coupled topology, avoiding low-frequency cutoff and thermal noise introduced by the DC blocking capacitor. The specific circuit structure is based on a non-inverting or inverting adder constructed from low-noise operational amplifiers. The inputs of the operational amplifiers are connected to the outputs of a DC reference voltage source and a precision attenuation network, respectively. To prevent leakage current from affecting the bias accuracy of high-impedance devices (such as MOSFET gates), an equipotential protection ring is arranged around the output traces of the signal superposition drive circuit. This protection ring is driven by the follower output of the operational amplifier, ensuring its potential is consistent with the signal line potential, thereby eliminating leakage current paths on the surface of the insulating substrate.

[0045] The output characteristics of the signal superposition driving circuit are as follows: the total voltage output to the control terminal of the semiconductor device is a linear weighted sum of the DC bias voltage and the attenuated AC excitation signal.

[0046] By adjusting the settings of the DC reference source and the amplitude of the waveform generated by the DAC, the system can independently control the static operating point of the device under test and the intensity of the injected excitation.

[0047] The main control and spectrum separation and de-embedding unit 400 generates excitation waveform data for system identification based on the data obtained during the reference noise detection stage.

[0048] The main control and spectrum separation / de-embedding unit 400 first defines the frequency grid within the measurement bandwidth. This is based on the sampling rate of the data acquisition unit 300. and the preset number of analysis points Establish frequency resolution The main control and spectrum separation / de-embedding unit 400 divides the discrete frequency points within the measurement bandwidth into a set of pilot frequencies. With noise observation frequency set Pilot frequency set The frequency selection method is as follows: in the low-frequency band (e.g., below 100Hz), a logarithmic distribution is used to match the 1 / f noise characteristics; in the mid-to-high frequency band, a linear, equally spaced distribution is used. (Noise observation frequency set) It includes all other frequency points except for the pilot frequency set.

[0049] After determining the frequency distribution, the main control and spectrum separation / de-embedding unit 400 performs adaptive amplitude calculation. The main control and spectrum separation / de-embedding unit 400 reads the estimated one-sided noise power spectral density. Calculate the amplitude of the excitation signal spectrum at each pilot frequency point. The calculation process includes two constraints: first, the signal-to-noise ratio constraint, which ensures that the pilot power is higher than the background noise; and second, the total power constraint, which ensures that the total signal after the superposition of all pilots is within the linear region of the device.

[0050] ; in: This indicates that the generated frequency domain excitation signal is at a frequency of The magnitude of the amplitude at that point; This represents the signal-to-noise ratio enhancement factor, typically ranging from 3 to 10. This represents the estimated one-sided noise power spectral density of the semiconductor device; This indicates that the frequency resolution is; This indicates the maximum total effective (RMS) voltage threshold that is allowed to be injected into the device under test; Represents the set of pilot frequencies The total number of valid pilots included; This represents the maximum allowable amplitude contribution of a single pilot under the assumption of equal power distribution.

[0051] For the two scenarios in the above formula, the main control and spectrum separation / de-embedding unit 400 performs the following logical judgment and assignment operations: The first scenario involves the current calculation frequency. Belongs to the pilot frequency set When the system executes The main control unit and the spectrum separation and de-embedding unit 400 simultaneously calculate the amplitude value based on the signal-to-noise ratio requirement and the amplitude average value based on the total power limit, and select the smaller of the two values ​​as the final excitation amplitude at that frequency point. This logic mechanism of taking the smaller value can simultaneously satisfy two boundary conditions: ensuring that the pilot signal is higher than the background noise to achieve effective measurement, and preventing the calculated drive amplitude from being too large in the high-noise frequency band, which could lead to device saturation or damage.

[0052] The second scenario involves the currently calculated frequency point. Not part of the pilot frequency set (i.e., belonging to the set of noise observation frequencies) When this occurs, the system performs a 0 value assignment operation. The main control and spectrum separation / de-embedding unit 400 forcibly assigns the amplitude at this frequency. Set to mathematical zero. This operation constructs the gaps between the teeth of a comb wave in the frequency domain, ensuring that there is no artificially injected excitation energy at these observation frequencies, thus providing an interference-free observation window for subsequent direct reading of the intrinsic noise of semiconductor devices.

[0053] After determining the amplitude spectrum, the main control and spectrum separation / de-embedding unit 400 performs phase encoding on the excitation signal to reduce the peak-to-average power ratio. The main control and spectrum separation / de-embedding unit 400 generates a set of pseudo-random phase sequences. The phase sequence remains constant during a single measurement and satisfies the condition that... Uniformly distributed within the interval. For the set of noise observation frequencies. At the frequency point in the frequency range, its phase value is set to zero.

[0054] After completing the frequency domain data construction, the main control and spectrum separation and de-embedding unit 400 transforms the complex frequency domain sequence into time domain waveform data through inverse discrete Fourier transform. Subsequently, the main control and spectrum separation / de-embedding unit 400 converts the floating-point time-domain waveform data... Mapped to integer input code values ​​of a digital-to-analog converter (DAC) The mapping process performs linear quantization based on the DAC's full-scale voltage range and bit depth, and adds a DC bias code value to adapt to the DAC's unipolar or bipolar output characteristics.

[0055] The generated integer input code value is written into the cyclic buffer of the precision bias and comb excitation unit 100. The main control and spectrum separation and de-embedding unit 400 controls the clock source of the digital-to-analog converter to remain in the same source or phase-locked with the sampling clock source of the data acquisition unit 300, ensuring that the waveform playback period length of the digital-to-analog converter is exactly equal to the FFT analysis frame length of the data acquisition unit 300, thereby eliminating the spectrum leakage caused by non-integer period sampling at the hardware physical level.

[0056] The adaptive analog equalization preamplifier unit 200 adopts a cascaded topology structure, which is divided into a first-stage fixed-gain low-noise amplifier stage and a second-stage programmable shaping amplifier stage in terms of physical circuit.

[0057] The first-stage fixed-gain low-noise amplifier is located at the very beginning of the signal chain and is designed as a high-input-impedance differential voltage amplifier structure. The input pairs of this stage consist of multiple pairs of low-noise N-channel junction field-effect transistors (JFETs) connected in parallel. The drains and sources of the multiple JFETs are interconnected to increase the equivalent transconductance and reduce the equivalent input voltage noise; the noise reduction is proportional to the square root of the number of parallel pairs. These JFET pairs are configured in a differential common-source topology and driven by a tail current circuit biased by a low-noise constant current source to provide suppression of power supply noise and common-mode interference.

[0058] The first-stage circuit is configured in fixed-gain mode, with its voltage gain set to a constant high gain value (e.g., 1000 times or 60 dB). This gain value is set according to the cascaded system noise theory (Friis formula), which states that by giving the first stage a sufficiently high gain, the equivalent input noise contribution of subsequent stages is divided by the first-stage gain, thus making its impact on the overall system noise figure negligible.

[0059] The output of the first-stage fixed-gain low-noise amplifier stage and the input of the second-stage programmable shaping amplifier stage are connected via an interstage AC coupling network. This interstage AC coupling network uses an RC high-pass filter composed of non-polar metallized polypropylene film capacitors and high-precision metal foil resistors. The cutoff frequency of this filter is set to one-tenth of the lower limit of the system measurement bandwidth (e.g., 1 MHz to 10 MHz) to completely block the DC drift voltage generated by the first-stage circuit and prevent it from entering the second stage and causing saturation cutoff at high gain.

[0060] The second-stage programmable shaping amplifier stage performs dynamic adjustment of the amplitude-frequency response. This stage is based on an inverting amplifier or instrumentation amplifier structure built from an ultra-low distortion operational amplifier. A high-precision metal foil input resistor is connected in series on the input path of this inverting amplifier. (For example One end of the resistor is connected to the output of the interstage AC coupling network, and the other end is connected to the inverting input of the operational amplifier. Its feedback loop integrates a programmable impedance network (PIN). This programmable impedance network consists of a precision resistor array, a capacitor array, and a low-leakage-current relay matrix (or a low-charge-injection analog switch).

[0061] The main control and spectrum separation / de-embedding unit 400 drives the relay matrix to engage and disengage by sending digital control words, thereby physically changing the resistance value in the feedback loop. and capacitance value Combinations. Through different and Whether connected in parallel or series, the frequency response curve of the second-stage circuit exhibits a pole-zero distribution characteristic opposite to the noise spectrum shape of the device under test, achieving attenuation of low-frequency high-amplitude noise and enhancement of high-frequency low-amplitude noise. The gain adjustment range of the second-stage circuit is designed to be -20dB to +40dB to accommodate the noise intensity differences of different types of semiconductor devices.

[0062] In terms of power supply, the first and second stage amplifier circuits are each powered by independent multi-stage low-dropout linear regulators (LDOs), and a common-mode choke is connected in series at the power input to suppress high-frequency interference on the power line. The circuit board layout adopts star grounding technology, with analog ground and digital ground connected at a single point at the power input to prevent backflow noise coupling.

[0063] The programmable impedance network is located in the negative feedback loop of the second-stage amplifier circuit of the adaptive analog equalization preamplifier unit 200. Physically, the programmable impedance network consists of a precision resistor array, a precision capacitor array, and a switch matrix. The precision resistor array and the precision capacitor array are connected in parallel.

[0064] A precision resistor array consists of multiple resistance values ​​weighted in binary (e.g.) The array consists of metal foil resistor branches arranged in a specific pattern. To prevent the operational amplifier output from saturating due to an open feedback loop at the moment of relay switching, a high-value resistor (e.g., 10Ω) is constantly connected in parallel across the precision resistor array. A fixed resistor is used as a backup path for DC feedback.

[0065] The precision capacitor array consists of multiple polypropylene film capacitor branches with binary-weighted capacitance values. The range of selectable capacitance values ​​covers... to , used to set different pole frequencies.

[0066] The switch matrix employs multiple dual-coil magnetic latching relays. The coil drive terminal of each relay is not directly connected to the main control unit, but rather via a Darlington transistor array (or a dedicated relay driver chip). A freewheeling diode is connected in reverse parallel across each relay coil to absorb the back electromotive force generated when the coil is de-energized, protecting the drive circuit. Magnetic latching relays were chosen because they consume energy only during state switching; no current flows through the coil in the latching state, thus eliminating thermoelectric interference. Furthermore, their high contact insulation resistance effectively blocks leakage current.

[0067] The main control and spectrum separation / de-embedding unit 400 controls the opening and closing of the switch matrix by sending serial or parallel digital signals to the drive circuit. All the resistor branches in the on state are connected in parallel to form an equivalent feedback resistance. All capacitor branches in the conducting state are connected in parallel to form an equivalent feedback capacitor. .

[0068] Complex impedance generated by programmable impedance network This determines the gain and bandwidth characteristics of the second-stage amplifier circuit. In the inverting amplifier configuration, the frequency response function of the second-stage circuit... Defined by the following complex number formula: ; in: This indicates that the second-stage amplifier circuit operates at a certain frequency. Voltage transfer function at the location; Indicates the frequency of the input signal; Represents the imaginary unit, satisfying ; Pi is a constant. This represents the input resistance of the second-stage amplifier circuit; The current equivalent total parallel resistance of the precision resistor array determines the amplifier's passband gain. ; This represents the current equivalent total parallel capacitance of the precision capacitor array; This represents the -3dB cutoff frequency determined by the feedback network.

[0069] Through independent adjustment and Based on these values, the system can perform two key functions: First, adaptive gain adjustment. The system adjusts the gain based on the estimated noise amplitude. By setting an appropriate amplification factor, the output signal amplitude can be fully utilized to ensure that the dynamic range of the data acquisition unit 300 is not lost, while clipping does not occur.

[0070] Second, dynamic bandwidth limitations. System adjustments. Set the cutoff frequency of the low-pass filter. Slightly higher than the current upper limit of the measurement focus band. This mechanism can effectively filter out broadband white noise and high-frequency interference outside the measurement band, preventing them from aliasing during data acquisition, thereby reducing the overall integrated noise floor of the system and improving the signal-to-noise ratio for measuring low-frequency weak signals.

[0071] After receiving the digital time-domain signal transmitted by the data acquisition unit 300, the main control and spectrum separation and de-embedding unit 400 performs frequency domain transformation and system identification process.

[0072] The main control and spectral separation and de-embedding unit captures 400 pairs of time-domain voltage sequences. Perform a Fast Fourier Transform (FFT). Before the transform, the system applies a rectangular window to the time-domain data, i.e., no weighted attenuation processing is performed. Since the sampling clock is hardware synchronized with the signal generation clock, and the sampling length is an integer multiple of the signal period, the rectangular window processing can maintain the amplitude and phase information of the orthogonal pilot signal without distortion, avoiding the energy diffusion introduced by conventional smoothing window functions.

[0073] The main control and spectrum separation / de-embedding unit 400, based on the pre-stored pilot frequency index, extracts data from the full-band complex spectrum. Extract the set of pilot frequencies A subset of the data. To improve the system's anti-interference capability, the main control and spectrum separation / de-embedding unit 400 employs a complex vector averaging algorithm. The system cumulatively reads... Pilot frequency point data for each measurement cycle ( Typically, values ​​of 4 to 16 are used to average their complex values, thereby suppressing the random background noise component at the pilot location by utilizing the incoherent nature of the noise before calculating the transfer function.

[0074] For each pilot frequency point The main control and spectrum separation / de-embedding unit 400 performs complex division to calculate the system's discrete frequency response at that frequency point. The discrete frequency response is calculated according to the following formula: ; in: Indicates the first The complex frequency response of the measurement system at each pilot frequency point; This indicates the received signal after vector averaging at a frequency of Complex values ​​at; This indicates that the known excitation signal set during the digital excitation generation stage is at a certain frequency. Complex values ​​at; Indicates belonging to the pilot frequency set The Pilot frequency points.

[0075] After obtaining the discrete frequency response sequence, the main control and spectrum separation / de-embedding unit 400 performs amplitude and phase separation processing on it to extract the amplitude and frequency response sequences respectively. and phase frequency response sequence For the phase frequency response sequence, the main control and spectrum separation / de-embedding unit 400 executes a phase unwinding algorithm. This algorithm detects the phase difference between adjacent pilot frequency points; when the absolute value of the phase difference exceeds... At that time, perform subsequent phase values The compensation operation eliminates the periodic truncation of the arctangent function, restoring the continuous phase curve that reflects the true group delay of the system. .

[0076] Based on the unwound amplitude and phase data, the main control and spectrum separation / de-embedding unit 400 constructs a continuous transfer function model covering the entire measurement bandwidth using a numerical interpolation algorithm. The system employs a cubic spline interpolation algorithm. In handling the interpolation boundaries, the system forcibly sets the gain at the DC frequency point (0Hz) to the theoretical product of the DC gains of the first and second stage amplifiers, and sets the phase shift to 0 or... This is to prevent the interpolation function from diverging at low frequencies.

[0077] The main control and spectrum separation / de-embedding unit 400 performs interpolation operations on the amplitude frequency response and the unwound phase frequency response, respectively, to calculate the noise observation frequency set. The estimated response value is calculated at each frequency point. Then, the interpolated amplitude and phase are recombined to generate the full-band interpolation transfer function. The construction of the full-band interpolation transfer function satisfies the following mapping relationship: ; in: Indicates at any frequency Estimating the system transfer function at the location; This indicates that the cubic spline interpolation function constructed based on discrete amplitude sequences has frequency... The calculated value at that location; This indicates that the cubic spline interpolation function constructed based on the unwound phase sequence has a frequency... The calculated value at that location; Represents the imaginary unit, satisfying ; It represents the base of the natural logarithm.

[0078] Through the above processing, the system reconstructs the system gain and phase shift characteristics at the noise observation frequency without applied excitation. This interpolation transfer function... As a complex calibration vector, it will be used in subsequent steps for frequency domain de-embedding of the noisy observation data.

[0079] The main control and spectrum separation / de-embedding unit 400 utilizes the full-band interpolation transfer function generated in the preceding steps. The raw frequency domain data stored in the receive buffer is processed to restore the intrinsic noise power spectral density of the device under test at the input port.

[0080] The main control and spectrum separation / de-embedding unit 400 performs a spectrum masking operation. The main control and spectrum separation / de-embedding unit 400 reads the system's preset pilot frequency set. and noise observation frequency set The index list. For the original frequency domain complex spectrum. The processor generates a valid subset of data, which contains only frequencies belonging to... The system uses logical judgment to determine which frequency indexes belong to the spectral components. The spectral data is invalidated or discarded directly. This operation eliminates frequency points containing high-amplitude excitation signals in the frequency domain, preventing the energy of the excitation signal from being incorrectly included in the device's noise power.

[0081] After extracting the complex spectrum data containing only the noise observation frequencies, the main control and spectrum separation / de-embedding unit 400 performs dewhitening and background noise subtraction operations. Since the measured noise data includes the noise of the device under test and the inherent noise of the measurement system itself (including voltage noise, current noise, and ADC quantization noise of the preamplifier), the system must perform background subtraction in order to obtain high-precision intrinsic device noise.

[0082] The main control and spectrum separation / de-embedding unit 400 first calculates the total noise power spectral density referred to the input, and then subtracts the system background noise spectrum that has been measured and stored in the input short-circuit or input open-circuit calibration mode. The calculation of the intrinsic noise power spectral density (PSD) of the device under test follows the physical relationship: ; in: This represents the intrinsic noise power spectral density of the semiconductor device under test referred to the input terminal, in units of... ; Represents the set of noise observation frequencies frequency in The original complex spectrum value measured at the location; This represents the system's full-band complex transfer function constructed by the interpolation algorithm at that frequency. The value at; This represents the pre-stored equivalent background noise power spectral density at the input of the measurement system; This represents the normalization factor for the equivalent noise bandwidth (ENBW) of the FFT window function (this value is 1 when a rectangular window is used). This represents the modulus operation for complex numbers.

[0083] If the calculation result If a negative value occurs (due to random fluctuations), the system clamps it to a smaller positive value or zero.

[0084] Through the above calculations, the system obtains the discrete frequency set The actual noise power spectral density is obtained. Since the pilot frequency data is removed during the spectral masking stage, the resulting noise spectrum curve is discontinuous on the frequency axis. To provide the user with a continuous and complete noise spectrum curve, the main control and spectrum separation / de-embedding unit 400 performs data completion operations.

[0085] The main control and spectrum separation / de-embedding unit 400 uses a frequency domain local linear interpolation algorithm to fill these gaps. For each removed pilot frequency point... The system selects the nearest effective noise frequency point on its left. and the effective noise frequency on the right The system calculates the weighted average or arithmetic mean of the noise values ​​at these two adjacent frequency points as an estimate of the noise power spectral density at the pilot frequency. .

[0086] When the pilot points are distributed at a single point, the calculation logic for data completion is as follows: ; in: Indicates the pilot frequency of the masked signal. The estimated noise power spectral density at the location; Indicates belonging to the pilot frequency set any frequency point; and This indicates the effective noise observation point immediately adjacent to the pilot point; and This represents the intrinsic noise power spectral density of the device calculated at adjacent frequency points.

[0087] After data completion processing, the discrete noise data points are reconstructed into a noise power spectral density curve with continuous frequency within the measurement bandwidth. This curve eliminates interference from the excitation signal, corrects the frequency response error of the measurement system, and removes the influence of the system's inherent noise. For the storage and management of frequency domain data and basic arithmetic averaging operations, those skilled in the art can use conventional array manipulation instructions or digital signal processing library functions, which are well-known techniques in the field and will not be elaborated upon here.

[0088] Specific application examples: Scenario Deployment: This example is applied to a wafer-level reliability testing laboratory in a semiconductor manufacturing plant to perform large-scale low-frequency noise parameter extraction on the input stage transistor of a novel low-noise operational amplifier (Op-Amp).

[0089] Hardware configuration: Test main station: 1 set of low-noise preamplifier semiconductor low-frequency noise test system.

[0090] Probe station: 1 semi-automatic shielded probe station with an integrated precision temperature control chuck (temperature control 25℃±0.1℃).

[0091] Test object: 12-inch wafer containing 500 N-channel JFET device structures under test.

[0092] Connection medium: low capacitance triaxial shielded probe and low noise shielded cable.

[0093] Production flow and data interaction process Step 1: The static bias and contact integrity verification probe station is moved to the Nth die position, and the probe is inserted.

[0094] Bias setting: The main control and spectrum separation / de-embedding unit 400 sends commands to the precision bias and comb excitation unit 100 to set the bias. .

[0095] Sensing and Verification: The data acquisition unit 300 reads the DC output level of the adaptive analog equalization preamplifier unit 200. The system detects the static drain current. .

[0096] Logical decision: If the measured current deviation exceeds the preset value (e.g.) Indicates opening a path. (Indicating a short circuit), the main control unit determines that there is poor contact or the device is damaged, skips the die directly and records the error code, and does not perform subsequent noise measurements to protect the precision measurement circuit.

[0097] Step 2: After the adaptive whitening and pilot injection contact verification are passed, the system enters the dynamic measurement mode.

[0098] Dynamic balancing: The system performs a reference noise detection. The main control unit and the spectrum separation and de-embedding unit 400 found that the device's low-frequency (1Hz) noise amplitude was as high as... The high frequency (100kHz) is close to the background noise.

[0099] Parameter configuration: The system automatically calculates and configures the adaptive analog equalization preamplifier unit 200 to high-pass filter mode (cutoff frequency set to 100Hz), while the precision bias and comb excitation unit 100 injects a comb pilot signal containing 20 frequency points.

[0100] Signal conditioning: After pre-whitening processing, the peak factor of the mixed signal waveform of the input data acquisition unit 300 is reduced from 5.0 to 1.8, and the dynamic range utilization of the ADC is increased from 15% to 92%.

[0101] Step 3: Spectral Separation, De-embedding, and Defect Attribution After 10 seconds of continuous data acquisition, the main control unit and the spectrum separation and de-embedding unit 400 perform data processing.

[0102] De-embedding computation: Calculation using the response of comb pilots The system deducts the high-frequency signal attenuation caused by cable capacitance (approximately 150pF) and corrects the amplitude-frequency characteristics of the preamplifier filter.

[0103] Defect identification: The system reconstructs the true noise spectrum at the device input. The system detected an abnormal Lorentz-type bulge in the 10Hz to 100Hz range, rather than a standard one. straight line.

[0104] Attribution analysis: Based on the turning frequency of the Lorentz component, the system determines that there is a specific deep-level defect (GR noise center) in the wafer region and marks the die as a reliability risk item.

[0105] Experimental verification and effect comparison To verify the actual effect of the system, a 5-day comparative experiment was conducted in the aforementioned laboratory, where 200 JFET devices produced in the same batch were grouped and tested.

[0106] Experimental Groups: Control group: A conventional fixed-gain low-noise amplifier (LNA, fixed gain 60dB) was used in conjunction with a dynamic signal analyzer. No real-time pilot injection was performed; only open-circuit / short-circuit calibration was performed before measurement.

[0107] Experimental group: The low-noise preamplifier semiconductor low-frequency noise test system of the present invention was activated, and the adaptive analog equalization and real-time comb pilot de-embedding functions were enabled.

[0108] Experimental data presentation: See attached document Figure 3 Chart description: The horizontal axis represents the measurement frequency (logarithmic coordinates, 1Hz-100000Hz), and the vertical axis represents the probability density distribution width of the ADC output code value (effective dynamic range).

[0109] Data Interpretation: The solid line (experimental group) in the graph maintained a high dynamic range utilization across the entire frequency band, especially in the low-frequency band, due to the removal of [unclear text - likely referring to a specific frequency range or parameter]. With the suppression of large signals, the ADC's ability to resolve weak signals remains constant. In the dashed area (control group), the overall gain has to be reduced at the low-frequency end to prevent clipping, resulting in smaller signal amplitude in the high-frequency range (>10000Hz), which is submerged in quantization noise, and the effective dynamic range shrinks sharply.

[0110] See attached document Figure 4 Chart description: The horizontal axis represents the sample number (Sample1-Sample200), and the vertical axis represents the standard deviation of high-frequency (100000Hz) thermal noise measurement.

[0111] Data Interpretation: Solid line (experimental group): The measured values ​​are closely convergent with a small standard deviation, indicating that the system has high consistency and repeatability. Dashed line (control group): The measured values ​​have large dispersion, which is due to the uncompensated changes in the transfer function introduced by small changes in probe contact resistance and cable position variations.

[0112] Summary of Results Comparison Conclusion: Experimental results show that this invention solves the problems in traditional testing through adaptive simulation equalization technology. The conflict between noise and broadband thermal noise in terms of dynamic range is addressed by employing a comb-type pilot real-time de-embedding technique to eliminate the influence of cable and interface impedance on high-frequency measurements. Based on this, a closed-loop measurement system is constructed that can characterize the low-frequency noise characteristics of semiconductor devices with higher accuracy, wider bandwidth, and faster speed, improving the efficiency and accuracy of chip reliability screening.

Claims

1. A low-noise preamplifier semiconductor low-frequency noise testing system, characterized in that, include: A precision bias and comb excitation unit (100) responds to a bias command by outputting a DC bias voltage, determining the static operating point of the semiconductor device, and superimposing a frequency-domain sparse comb excitation signal generated based on waveform synthesis data onto the DC bias voltage and injecting it into the semiconductor device. An adaptive analog equalization preamplifier unit (200) receives a mixed output signal containing device intrinsic noise and the response to the comb excitation signal, and performs analog domain spectral shaping on the mixed output signal through an internal analog filter network with adjustable frequency response characteristics. The data acquisition unit (300) converts the shaped mixed output signal into a digital time-domain sequence; The main control and spectrum separation and de-embedding unit (400) sends bias commands and waveform synthesis data to the precision bias and comb excitation unit (100) and receives the digital time-domain sequence transmitted by the data acquisition unit (300); the main control and spectrum separation and de-embedding unit (400) calculates the system transfer function based on the response amplitude of the comb excitation signal at the pilot frequency point, and performs de-embedding processing on the noise data at non-pilot frequency points to calculate the input noise power spectrum of the semiconductor device.

2. The low-noise preamplifier semiconductor low-frequency noise testing system according to claim 1, characterized in that, The precision bias and comb excitation unit (100) includes, in its physical circuit architecture, a digital isolation interface, a dynamic excitation generation module, a precision attenuation network, and a signal superposition driving circuit connected in sequence, and is independently equipped with a DC reference voltage source. The dynamic excitation generation module converts the waveform synthesis data into an analog stepped wave, and outputs the original AC excitation signal after filtering. The precision attenuation network attenuates the amplitude of the original AC excitation signal; The signal superposition driving circuit adopts a fully DC-coupled topology. The input terminal of the signal superposition driving circuit is connected to the DC reference voltage source and the precision attenuation network respectively, and linearly synthesizes the DC bias voltage and the attenuated AC excitation signal.

3. The low-noise preamplifier semiconductor low-frequency noise testing system according to claim 1, characterized in that, The main control and spectrum separation and de-embedding unit (400) divides the frequencies within the measurement bandwidth into a set of pilot frequencies and a set of noise observation frequencies; When generating the comb-shaped excitation signal, the main control and spectrum separation and de-embedding unit (400) sets the amplitude of the frequency points belonging to the pilot frequency set to a non-zero value. The non-zero value is the smaller of the amplitude value calculated based on the signal-to-noise ratio requirement and the amplitude average value calculated based on the total power limit. The amplitude of frequency points belonging to the noise observation frequency set is forcibly set to zero.

4. The low-noise preamplifier semiconductor low-frequency noise testing system according to claim 1, characterized in that, The adaptive analog equalization preamplifier unit (200) adopts a cascaded topology, including a first-stage fixed-gain low-noise amplifier stage, an inter-stage AC coupling network, and a second-stage programmable shaping amplifier stage; The first fixed-gain low-noise amplifier stage includes an input pair of transistors, which consists of a differential common-source topology composed of multiple pairs of parallel low-noise junction field-effect transistors. The second-stage programmable shaping amplifier stage is located after the interstage AC coupling network, and the feedback loop of the second-stage programmable shaping amplifier stage integrates a programmable impedance network. The main control and spectrum separation and de-embedding unit (400) adjusts the programmable impedance network according to the estimated noise power spectral density, and the second-stage programmable shaping amplifier stage has frequency response characteristics that are opposite to the noise spectrum shape of the semiconductor device.

5. A low-noise preamplifier semiconductor low-frequency noise testing system according to claim 4, characterized in that, The programmable impedance network consists of a precision resistor array, a capacitor array, and a switch matrix; The precision resistor array and the precision capacitor array are connected in parallel, and the switch matrix includes multiple relays; The main control and spectrum separation and de-embedding unit (400) changes the equivalent parallel total resistance and equivalent parallel total capacitance in the feedback loop by controlling the closing and opening of the switch matrix.

6. The low-noise preamplifier semiconductor low-frequency noise testing system according to claim 1, characterized in that, The main control and spectrum separation and de-embedding unit (400) performs interpolation fitting of the system transfer function; Specifically, this includes: extracting the discrete system frequency response at the pilot frequency point, and performing amplitude-phase separation and phase dewinding processing on the discrete system frequency response; The cubic spline interpolation algorithm is used to perform interpolation operations on the amplitude frequency response and the phase frequency response after unwinding, respectively, to generate a continuous full-band interpolation transfer function covering the entire measurement bandwidth.

7. A low-noise preamplifier semiconductor low-frequency noise testing system according to claim 6, characterized in that, When performing the de-embedding process, the main control and spectrum separation and de-embedding unit (400) performs a spectrum masking operation to remove pilot frequency point data containing comb excitation signal energy and retain noise observation frequency point data. The full-band interpolation transfer function is then used to perform a reverse operation on the noise observation frequency point data, subtracting the pre-stored system background noise power spectral density.

8. A low-noise preamplifier semiconductor low-frequency noise testing system according to claim 7, characterized in that, The main control and spectrum separation and de-embedding unit (400) performs frequency domain data completion operation; For the pilot frequency points that are eliminated by the spectrum mask operation, the main control and spectrum separation and de-embedding unit (400) selects the power spectral density values ​​of the left and right noise observation frequency points that are adjacent to the pilot frequency points, and calculates the weighted average or arithmetic average as the noise power spectral density estimate at the pilot frequency.

9. A low-noise preamplifier semiconductor low-frequency noise testing system according to claim 1, characterized in that, The precision bias and comb excitation unit (100) includes a digital-to-analog converter, and the waveform playback clock source of the digital-to-analog converter is kept in the same source or phase-locked with the sampling clock source of the data acquisition unit (300). The main control and spectrum separation and de-embedding unit (400) sets the playback period length of the comb excitation signal to be equal to the fast Fourier transform analysis frame length of the data acquisition unit (300), and applies rectangular window processing to the acquired time-domain data.

10. A low-noise preamplifier semiconductor low-frequency noise testing system according to claim 2, characterized in that, The signal superposition driving circuit includes an operational amplifier, and an equipotential protection ring is arranged around the output traces of the signal superposition driving circuit. The equipotential protection ring is connected to the follower output terminal of the operational amplifier, and the potential of the equipotential protection ring is consistent with the potential of the output signal line.

Images