Overview
Sirma has partnered with Resonant Inc*, a NASDAQ-listed company that specializes in radio frequency (RF) filter technology for mobile devices. Together, they designed and deployed a cloud-based machine learning platform that revolutionizes how semiconductor fabrication facilities (FABs) measure and analyze wafers during manufacturing.
This platform utilizes proprietary machine learning techniques to conduct simple, non-destructive wafer-probe measurements of Process Control Monitors (PCMs), providing Critical to Quality (CTQ) dimensional data across the entire wafer. By replacing expensive and destructive legacy imaging methods, Sirma’s solution has given Resonant and its FAB partners a significant competitive advantage. They now experience faster time to market, enhanced yield optimization, and measurable reductions in production losses.
Challenge
Semiconductor manufacturing requires extreme precision, and traditional methods such as Focused Ion Beam Scanning Electron Microscopy (FIB-SEM) and Transmission Electron Microscopy (TEM) have limitations: they are slow, costly, destructive, and only measure specific locations on wafers, leaving significant blind spots. This inefficiency leads to a 6 to 12 month delay in bringing new FAB processes to revenue and an additional 2 to 3 months for yield optimization. Manufacturers lose about 3% of their annual yield due to undetected issues. Resonant sought a faster, non-destructive, and cost-effective approach that provides full-wafer coverage instead of isolated measurements.
Project Scope
Sirma developed a cloud-based wafer analysis platform that includes a complete analytical pipeline. Key aspects of the project involved creating machine learning algorithms that are resilient against measurement parasitics for effective use in semiconductor fabrication environments. The team also implemented logic to extract critical-to-quality dimensional data from standard wafer-probe measurements and integrated full-wafer spatial coverage mapping, replacing single-point measurements from FIB-SEM and TEM. All proprietary ML techniques were secured as trade-secret algorithms within the cloud infrastructure. Additionally, data pipelines were established to connect wafer-probe instruments to the cloud for seamless automated data ingestion and analysis at production scale.
Solution
The platform leverages a key insight: dimensional information traditionally obtained through destructive imaging can be accurately inferred from standard wafer-probe measurements of PCMs, given that the machine learning models are designed to mitigate measurement parasitics. Sirma’s proprietary algorithms were developed to correlate PCM electrical measurements with known physical dimensions, allowing them to extract critical dimensional data from existing probe measurements without needing additional hardware, destructive processes, or time delays in production. The cloud deployment enables FABs to stream measurement data and perform extensive dimensional analysis across the entire wafer surface while maintaining secure access to the algorithms and protecting Resonant’s intellectual property.
Results
The platform delivered quantified, operational impact across every stage of the semiconductor manufacturing lifecycle:
- 6 to 12 months cut from time to market when bootstrapping each new FAB process, directly accelerating revenue generation from new process nodes;
- Approximately 6 months faster new product introductions, compressing the development cycle from process qualification to customer shipment;
- 2 to 3 months reduction in yield optimization time between production ramp and full mass production;
- 3% yield saved per year by eliminating excursions and out-of-control production incidents through continuous, full-wafer monitoring;
- Full-wafer coverage replacing the sparse, location-limited data provided by FIB-SEM and TEM, giving process engineers a complete picture of dimensional variation across every wafer;
- Non-destructive analysis preserving wafers that would otherwise be sacrificed for measurement, reducing material waste and cost.
Technologies We Used
- Machine Learning - proprietary supervised ML algorithms; regression and dimensional inference models; parasitic-insensitive signal processing;
- Data Engineering - wafer-probe PCM measurement ingestion pipelines; full-wafer spatial data mapping and interpolation;
- Cloud Platform - secure cloud deployment architecture; multi-tenant FAB data isolation; trade-secret IP protection layer;
- Measurement Integration - PCM wafer-probe instrument connectors; electrical-to-dimensional data transformation;
- Analytics & Reporting - CTQ dimensional output generation; yield excursion detection and alerting; process control monitoring dashboards;
- Security - proprietary algorithm encapsulation; encrypted data transmission; access-controlled FAB partner environments.
Sirma’s Partnership with the client
Sirma’s collaboration with Resonant Inc. focused on transforming a long-stagnant measurement paradigm. Resonant contributed expertise in RF filter design and semiconductor engineering to address costly measurement limitations, while Sirma provided machine learning algorithms and cloud architecture to create a production-ready platform. The partnership involved developing robust algorithms suitable for the noisy environment of FAB production floors. The final product protects Resonant’s competitive advantage and delivers measurable returns to FABs using it, showcasing Sirma’s commitment to transformative solutions in the industry.
Disclaimer: Murata Manufacturing Co., Ltd. acquired Resonant Inc. on March 28, 2022.
