Keysight Technologies has launched the FITS-8CH platform, a solution designed to support the reliability of high-speed interconnects in AI data centres. The platform assists manufacturers in assessing bit error ratio (BER) and forward error correction (FEC) performance across network architectures for 400GE, 800GE, and emerging 1.6T Ethernet networks.
Increasing data throughput demands and the complexity of AI workloads require reliable network systems. Even minor error rates can affect performance at scale. The FITS-8CH platform enables digital-layer validation across complete interconnect assemblies, allowing engineers to detect potential reliability issues during design, development, and manufacturing stages.
The platform validates error performance across all relevant electrical lane speeds, including PAM4 signals at 53 Gb/s, 106 Gb/s, and 212 Gb/s, and complements Keysight’s existing physical-layer testing solutions. It is intended to support chip, optical, and copper interconnect manufacturers in ensuring reliability from design through production.
Key features include:
- Comprehensive Validation: Multilane BER and FEC testing across all channels for accurate performance assessment.
- Flexible Architecture: Supports multiple electrical connection configurations without major redesigns.
- High-Quality Signal Integrity: Generates IEEE-compliant signals for precise BER and FEC measurements.
- Automated Tuning: Adjusts signal outputs per lane to improve measurement consistency and production-readiness projections.
- Proactive Detection: Identifies potential issues, such as mechanical misalignments or thermal failures, early in the process to reduce production risks.
With AI data centres playing a critical role in digital infrastructure, ensuring interconnects are validated under real-world conditions is essential. Keysight’s FITS-8CH platform aims to provide a structured approach to error performance, helping manufacturers reduce risk, maintain quality, and meet the requirements of next-generation high-speed networks.