Machine Learning IP Recommendation System for Semiconductor Design
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The selection of suitable silicon intellectual properties (IPs) for product design is time-consuming due to the difficulty in capturing their features from various formats and the lack of confidence, especially for new products in the semiconductor industry, where growing IP portfolios require extensive searching with limited technical support.
Innovation Solution
A machine learning-based IP recommending system that retrieves and analyzes IP portfolios, usage data, and product data to train a model that predicts and recommends relevant IPs for a desired product design, utilizing a graphical user interface for user input and displaying sorted results based on usage probability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual searching and analysis of IP portfolios is performed, then comprehensive IP selection is possible, but the process becomes extremely time-consuming
Solution Approach 1:
The patent replaces manual mechanical searching and analysis with an automated machine learning system. The ML model automatically retrieves IP portfolios, extracts features from various document formats, trains on usage data, and predicts suitable IPs, eliminating the need for manual review of numerous IP documents while maintaining or improving selection accuracy
Solution Approach 2:
The patent introduces a machine learning model as an intermediary between the IP portfolio database and the user. This intermediary automatically processes the complex task of matching product requirements with suitable IPs by learning from historical usage data, thereby reducing the time users would otherwise spend on manual searching and analysis
2Adaptability or versatility
If extensive IP portfolios are maintained for advanced technologies, then more IP options are available, but the difficulty of finding useful IPs increases
Solution Approach 1:
The patent replaces manual IP feature extraction and analysis with automated machine learning processes. The system automatically retrieves IP portfolios in various formats, extracts relevant features using ML techniques, and processes this information without manual intervention, making the growing volume of IP data manageable
Solution Approach 2:
The patent transforms IP portfolio data from unstructured or semi-structured formats into structured features that the ML model can process. By changing the representation parameters of IP data and training the model on usage patterns, the system makes it easier to detect and measure relevant IP characteristics across diverse formats
3Ease of operation
If manual IP analysis is performed without technical support, then independence is maintained, but the confidence in IP selection decreases
Solution Approach 1:
The patent replaces manual expert analysis with an automated machine learning system that provides consistent, data-driven recommendations. The system retrieves usage data, trains models on historical patterns, and generates predictions with confidence metrics, eliminating the need for manual technical support while improving selection confidence
Solution Approach 2:
The patent implements a feedback mechanism where the ML model is trained on historical IP usage data and continuously improves its predictions. The system provides confidence scores for its recommendations and can be refined based on user feedback, thereby increasing ease of operation and selection confidence without requiring external technical support
Data Source
AI summary
An intellectual property (IP) recommending method and an IP recommending system are provided. In the method, a plurality of IP portfolios respectively designated for a plurality of product designs are retrieved and usage data of a plurality of IPs included in each of the plurality of IP portfolios are extracted. A machine learning (ML) model is trained by using a portion of the retrieved IP portfolios and the extracted usage data. In response to receiving at least one criterion for a desired product design from a user, a plurality of IPs adapted for the desired product design are predicted based on the ML model and recommended for the user.