CN-122027017-A - InP optical communication laser chip reliability prediction and aging screening method

CN122027017ACN 122027017 ACN122027017 ACN 122027017ACN-122027017-A

Abstract

The invention provides a method for predicting the reliability and screening aging of an InP optical communication laser chip, and relates to the technical field of semiconductor laser manufacturing and reliability assessment. The method comprises the steps of inducing transient thermal lens effect through asymmetric bipolar current perturbation, removing charge interference through reverse bias, capturing dynamic resistance and wavelength chirp sequence reflecting lattice dislocation climbing, constructing a physical failure mode database, mapping degradation characteristics to an epitaxial stress relaxation and non-radiation composite mechanism through frequency domain stripping intrinsic thermal damping to achieve defect fingerprint deep analysis, and finally carrying out service life prediction by taking a carrier rate equation and an entropy augmentation law as constraints based on a physical embedded deep learning architecture, and evaluating residual lattice structure margin by utilizing a second derivative of a prediction track. The method solves the problems of insufficient hidden defect identification and prediction against the physical rule, and realizes high-precision performance grading and risk interception.

Inventors

ZHOU SHAOFENG
DING LIANG
Lv tianjian
FANG ZIXUN
LUO JUNBO

Assignees

深圳市星汉激光科技股份有限公司

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (8)

1. The method for predicting the reliability and screening the aging of the InP optical communication laser chip is characterized by comprising the following steps of: Sp1, performing an accelerated aging test, namely applying stepped incremental injection current stress and high-frequency pulse perturbation to an indium phosphide optical communication laser chip on a constant high-temperature base, and collecting a transient photoelectric response parameter sequence of the chip in the accelerated aging test process in real time, wherein the transient photoelectric response parameter sequence at least comprises a threshold current drift amount, a carrier composite heat dissipation rate and a luminescence wavelength red shift amount; Sp2, constructing a failure mode database, extracting nonlinear degradation characteristic components in a transient photoelectric response parameter sequence, mapping the nonlinear degradation characteristic components with a bottom layer physical lattice defect mechanism of an InGaAs phosphorus multiple quantum well active region, wherein the bottom layer physical lattice defect mechanism comprises epitaxial stress relaxation, dark line defect proliferation and cleavage plane amorphization, and constructing the failure mode database covering various physical degradation tracks based on the mapping relation; Sp3, predicting life trend through a data analysis model, establishing a data analysis model fused with a semiconductor carrier depletion physical law and a deep neural network, inputting the extracted nonlinear degradation characteristic component into the data analysis model, and deducing an optical power attenuation intrinsic track of the indium phosphide optical communication laser chip under standard working conditions by combining a reference standard in a failure mode database, so as to predict the life trend of the chip reaching a catastrophic optical damage critical point; Sp4, realizing reliability screening and performance grading before delivery, and executing automatic reliability screening and performance grading on the chip subjected to accelerated aging test based on a life trend prediction result and a residual lattice structure margin output by the data analysis model, and outputting a high-reliability backbone network communication grade, a conventional access network communication grade and a lattice defect rejection grade.
2. The method for predicting reliability and screening aging of an InP optical communication laser chip according to claim 1, wherein the gradient increasing injection current stress in Sp1 is overlapped with high-frequency pulse perturbation, and the high-frequency pulse perturbation is used for exciting bound charges of deep trap energy levels in an InGaAsP multiple quantum well active region to induce a transient space hole burning effect in the chip so as to accelerate exposure of epitaxial growth lattice dislocation defects in an early dormant state.
3. The method for predicting reliability and screening aging of an InP optical communication laser chip according to claim 1, wherein the carrier recombination heat dissipation rate in Sp1 is obtained by separating a Joule thermal damping component and an Auger recombination electric damping component in a chip dynamic micro-junction resistance sequence, and is used for separately characterizing an abnormal local temperature rise gradient caused by the sharp increase of a non-radiative recombination center in an active region.
4. The method for predicting reliability and screening aging of an InP optical communication laser chip according to claim 1, wherein the step of mapping nonlinear degradation characteristic components to underlying physical lattice defect mechanisms in Sp2 comprises calculating a spectral frequency domain broadening coefficient caused by microscopic mismatch of an epitaxial lattice constant, and storing a step change of the spectral frequency domain broadening coefficient as a determination fingerprint for identifying an epitaxial stress relaxation failure mode in a failure mode database.
5. The method for predicting reliability and screening aging of an InP optical communication laser chip according to claim 1, wherein a physical energy conservation penalty term consisting of spontaneous radiation coefficients and stimulated radiation sectional areas is rigidly embedded in a cost function of network weight back propagation optimization of a data analysis model in Sp3, and life trend of output of the constraint data analysis model is required to strictly follow a semiconductor thermodynamic entropy increasing rule.
6. The method for predicting reliability and screening aging of an InP optical communication laser chip according to claim 1, wherein the Sp3 data analysis model further utilizes a Bayesian probabilistic inference network to quantitatively analyze uncertainty of a predicted lifetime trend, and outputs a time span mean value of the predicted lifetime and a continuous confidence interval boundary.
7. The method for predicting reliability and screening aging of an InP optical communications laser chip according to claim 1, wherein the condition for determining the communication level of the highly reliable backbone network in Sp4 is that the predicted optical power attenuation eigen track exhibits a linear gradual degradation characteristic within a specified period, and the attenuation rate of the remaining lattice structure margin is continuously lower than a preset safety baseline.
8. The method for predicting reliability and screening aging of an InP optical communication laser chip according to claim 1, wherein the method further comprises a closed loop feedback optimization step, wherein the dynamic calibration and the iterative update of network weights are performed on a failure mode database in Sp2 and a data analysis model in Sp3 by using a field adaptive migration learning algorithm through collecting actual service degradation data of a shipped chip in a client long-distance optical fiber transmission device.

Description

InP optical communication laser chip reliability prediction and aging screening method Technical Field The invention relates to the technical field of semiconductor laser manufacturing and photoelectronic device reliability evaluation, in particular to a method for predicting the reliability and screening aging of an InP optical communication laser chip. Background Along with the evolution of the global optical communication network to ultra-high speed and high integration, the stability of the backbone network and the data center communication system is directly determined by the reliability of the long-term operation of the indium phosphide-based distributed feedback laser chip serving as a core light source of the optical module. In the existing laser chip production flow, reliability screening before delivery mainly depends on static accelerated aging test. Conventional accelerated burn-in testing typically places the chip in a constant high temperature environment and applies a dc injection that is several times the operating current. The screening logic is mainly based on photoelectric parameter comparison before and after aging, and whether the chip is qualified or not is judged by monitoring the variation of threshold current or the attenuation proportion of slope efficiency after aging for a period of time. And when the parameter fluctuation exceeds a preset fixed threshold value, the chip is regarded as having failure risk and is rejected. In addition, the existing data analysis means mostly adopt a linear regression based on statistics or a simple black box deep learning model, and the life trend of the chip is fitted by collecting a large number of power degradation samples. In an application scenario for high-density wavelength division multiplexing and high-capacity coherent optical communication, the above-mentioned prior art has the following significant limitations: First, conventional static burn-in tests are not effective in exciting and identifying deep hidden defects in indium phosphide materials in a short period of time. Many dislocation climbs or proliferation of non-radiative recombination centers at interfaces due to uneven epitaxial growth stress often show parameter stabilization artifacts in the early stage of aging, but catastrophic optical damage can occur thousands of hours after actual service, leading to sudden failure of the backbone network. Second, existing data analysis models lack constraints on the physical nature of semiconductors. The pure data fitting-based model is extremely easy to fall into an overfitting state when processing indium phosphide chip data with small samples and multiple parameter fluctuation, and normal process fluctuation and potential physical degradation cannot be accurately distinguished. The life trend predicted by the model often violates the physical rule of carrier transportation and thermodynamic entropy increase, so that the prediction accuracy cannot meet the requirements of carrier-class equipment. Finally, existing screening schemes lack a sophisticated performance ranking strategy. Because the residual lattice structure margin of the chip in the whole life cycle cannot be quantitatively evaluated, a manufacturer often adopts a cut-to-cut screening standard. The method not only leads high-quality chips with high reliability potential not to be accurately identified and applied to long-distance backbone networks, but also leads partial chips with standard performance in a short period and extremely high long-term risks to be mixed into a supply chain, and seriously influences the hierarchical deployment and cost optimization of the optical communication system. Therefore, a screening scheme which can go deep into the physical failure mechanism of the bottom layer of the indium phosphide chip, combine the constraint of the physical law and realize the precise life prediction and multistage classification is needed, so as to solve the technical problems that the factory screening precision of the optical communication laser chip is insufficient and the reliability cannot be predicted in a closed loop. Disclosure of Invention Technical problem to be solved Aiming at the defects of the prior art, the invention provides a method for predicting the reliability and screening aging of an InP optical communication laser chip, which solves the following problems: 1. The direct current steady state aging adopted in the prior art is difficult to excite deep level defects in the indium phosphide material. The invention utilizes strong reverse bias to effectively empty trap charges in the quantum well by the excitation of asymmetric bipolar pulse, so that dislocation climbing and lattice distortion in a dormant state are forced to be displayed in advance in a screening stage. By monitoring the transient wavelength chirp and the micro-fluctuation of the micro-junction resistance, the extremely early capture of deep physical damage such as dark line d