Chen Leilei

Bio: Chen Leilei is an academic researcher from Sichuan University of Science and Engineering. The author has contributed to research in topics: Hyperparameter optimization & Support vector machine. The author has co-authored 1 publications.

Signal Integrity Analysis Based on SVR Improved Algorithm

Abstract: Compared with the traditional support vector machine regression (SVR), the SVR hyperparameter fast optimization algorithm can improve the accuracy of the prediction results. However, the data shows that when the training sample is too large, it will increase the complexity of model learning, resulting in too long modeling time. Therefore, we refer to the most effective support vector set search method in the variable selection and sparse support vector machine (VSߝSSVM) algorithm, and appropriately fit the “advantages” of these two algorithms to construct a fast optimization hyperparameter and sparse support vector machine (FOH-SSVM) algorithm. In this work, we use the algorithm to solve the problem of signal integrity. The experimental results show that the modeling time required by the FOH-SSVM algorithm is 1%, which greatly reduces the modeling time. At the same time, the prediction accuracy of the algorithm is increased by 8%, ensuring good prediction performance.

Artificial Intelligence based advanced Signal Integrity prediction

Abstract: As the signaling speeds continue to increase, maintaining Signal Integrity (SI) for the complete customer design space is a huge challenge. These constraints, along with the limitations of traditional methods of design space inclusion and channel behavior prediction pose significant risk to system design. Specific focus is needed on design space utilization techniques used for factoring in platform variability. Interfaces like PCIe Gen5/Gen6/Gen4 etc. exhibit higher order behaviors that can’t be modelled by current prediction algorithm like Response Surface Method (RSM). This leads to inaccurate system behavior understanding and results in unreliable platform design recommendations. To minimize design risk and achieve highly reliable scaling of Platform Design Guide (PDG) solution, this paper discusses the implementation of an Artificial Intelligence (AI) based methodology to cover complete design space and predict higher order system behaviors with high accuracy. Current SI method involves RSM type Design of Experiments (DOE) creation and results prediction using second order RSM as shown in Fig. 2(a). It has limitations since RSM uses only three variable levels therefore doesn’t cover the entire design space. It can only model up to second order system behavior. These issues can be addressed using proposed AI based methodology shown in Fig. 2(b). These AI techniques have been encapsulated into an AI based tool called Fitpro which fully automates space filled DOE creation and SI results prediction. Fitpro significantly reduces manual interventions and positively impacts efficiency.

Artificial Intelligence based advanced Signal Integrity prediction

Abstract: As the signaling speeds continue to increase, maintaining Signal Integrity (SI) for the complete customer design space is a huge challenge. These constraints, along with the limitations of traditional methods of design space inclusion and channel behavior prediction pose significant risk to system design. Specific focus is needed on design space utilization techniques used for factoring in platform variability. Interfaces like PCIe Gen5/Gen6/Gen4 etc. exhibit higher order behaviors that can’t be modelled by current prediction algorithm like Response Surface Method (RSM). This leads to inaccurate system behavior understanding and results in unreliable platform design recommendations. To minimize design risk and achieve highly reliable scaling of Platform Design Guide (PDG) solution, this paper discusses the implementation of an Artificial Intelligence (AI) based methodology to cover complete design space and predict higher order system behaviors with high accuracy. Current SI method involves RSM type Design of Experiments (DOE) creation and results prediction using second order RSM as shown in Fig. 2(a). It has limitations since RSM uses only three variable levels therefore doesn’t cover the entire design space. It can only model up to second order system behavior. These issues can be addressed using proposed AI based methodology shown in Fig. 2(b). These AI techniques have been encapsulated into an AI based tool called Fitpro which fully automates space filled DOE creation and SI results prediction. Fitpro significantly reduces manual interventions and positively impacts efficiency.

Artificial Intelligence based advanced Signal Integrity prediction

Abstract: As the signaling speeds continue to increase, main-taining Signal Integrity (SI) for the complete customer design space is a huge challenge. These constraints, along with the limitations of traditional methods of design space inclusion and channel behavior prediction pose significant risk to system design. Specific focus is needed on design space utilization techniques used for factoring in platform variability. Interfaces like PCIe Gen5/Gen6/Gen4 etc. exhibit higher order behaviors that current prediction algorithm like Response Surface Method (RSM) simply cannot model. This leads to inaccurate system behavior understanding and results in unreliable platform design recommendations. To minimize design risk and achieve highly reliable scaling of Platform Design Guide (PDG) solution, this paper discusses the implementation of an Artificial Intelligence (AI) based methodology to cover complete design space and predict higher order system behaviors with high accuracy. The goal is to achieve a model with at least 90 % R square and maximum 5% of result range Root Mean Square Error (RMSE). Current SI method of Design of Experiments (DOE) creation and results prediction consists of creating a combined RSM type DOE table and fitting it with second order RSM modelling in JMP as shown in Fig.2(a). It has limitations since RSM uses only three variable levels therefore doesn’t cover the entire design space. It can only model up to second order system behavior. These issues can be addressed using proposed AI based methodology which effectively captures the complete design variance using space filling algorithm shown in Fig.2(b). Paired with this, various AI based algorithms are explored for advanced SI results prediction. These techniques have been encapsulated into an AI based tool which supports automatic DOE creation and predicts the system behavior post simulation in a SINGLE ITERATION. This helps reduce manual interventions and improve efficiency along with highly desirable R square and RMSE values.

Artificial Intelligence based advanced Signal Integrity prediction

Abstract: As the signaling speeds continue to increase, main-taining Signal Integrity (SI) for the complete customer design space is a huge challenge. These constraints, along with the limitations of traditional methods of design space inclusion and channel behavior prediction pose significant risk to system design. Specific focus is needed on design space utilization techniques used for factoring in platform variability. Interfaces like PCIe Gen5/Gen6/Gen4 etc. exhibit higher order behaviors that current prediction algorithm like Response Surface Method (RSM) simply cannot model. This leads to inaccurate system behavior understanding and results in unreliable platform design recommendations. To minimize design risk and achieve highly reliable scaling of Platform Design Guide (PDG) solution, this paper discusses the implementation of an Artificial Intelligence (AI) based methodology to cover complete design space and predict higher order system behaviors with high accuracy. The goal is to achieve a model with at least 90 % R square and maximum 5% of result range Root Mean Square Error (RMSE). Current SI method of Design of Experiments (DOE) creation and results prediction consists of creating a combined RSM type DOE table and fitting it with second order RSM modelling in JMP as shown in Fig.2(a). It has limitations since RSM uses only three variable levels therefore doesn’t cover the entire design space. It can only model up to second order system behavior. These issues can be addressed using proposed AI based methodology which effectively captures the complete design variance using space filling algorithm shown in Fig.2(b). Paired with this, various AI based algorithms are explored for advanced SI results prediction. These techniques have been encapsulated into an AI based tool which supports automatic DOE creation and predicts the system behavior post simulation in a SINGLE ITERATION. This helps reduce manual interventions and improve efficiency along with highly desirable R square and RMSE values.