Key Features:
· 1 ns timing accuracy: up to 100 times better
· ±2 ppb/°C dF/dT (frequency slope as a function of temperature): up to 25 times better
· 6 × 10⁻¹² Allan deviation (ADEV): up to 8 times lower
· ±50 ppb frequency stability between -40 and 105 °C: up to 4 times better
· 3.2 mm × 2.5 mm (8 mm²) size: up to 2 times smaller
· Digital frequency tuning simplifies the design of timing-sensitive networks
· Eliminates dips and micro-skips inherent in quartz technology
· Resistant to shock, vibration, and plate flex
Availability:
The Elite 2 Super-TCXO is currently available as samples, and commercial production is expected to begin in the third quarter of 2026.
The Elite 2 Super-TCXO is available in 3.2 mm × 2.5 mm plastic and 5.0 mm × 3.2 mm ceramic packages. Part numbers: SiT5234, SiT5235, SiT5434, and SiT5435.
“Industry reports show that GPU utilization in AI clusters can be as low as 20 to 40 percent, placing a significant and largely hidden burden on AI infrastructure,” said Piyush Sevalia, Chief Commercial Officer at SiTime. “AI workloads are distributed across GPUs at strictly coordinated time slots. Even small synchronization errors force wait cycles to prevent data corruption and, in extreme cases, can lead to GPU timeouts and system restarts. Poor synchronization directly limits GPU utilization.”
Sevalia continued, “To address this, the industry is moving toward a 10-nanosecond time synchronization target across an entire AI cluster, up from the current microsecond. We collaborated closely with leading AI system architects from hyperscalers and silicon vendors and concluded that the right oscillator can significantly improve synchronization across the entire cluster. That’s why we developed the Elite 2 Super-TCXO.” The device offers sub-nanosecond synchronization, 10 times better than the target, thanks to its exceptional short-term thermal stability. With these features, the Elite 2 minimizes timing errors between GPUs, enabling greater system utilization, increased performance, and improved performance per watt. This is the result of SiTime's leadership and systems thinking applied to one of AI's most challenging problems.
“AI networks must operate with extremely high efficiency to fully leverage the expensive resources of GPUs,” said Sameh Boujelbene, vice president of Dell’Oro Group. “As AI back-end infrastructure is renewed at a much faster pace than traditional, non-accelerated infrastructure, precise timing is becoming increasingly critical to maintaining performance in rapidly evolving data center architectures.”.
