Elite RF Differential Super-TCXO

3X Better Synchronization 

As the adoption of AI continues at an extraordinary pace, datacenter utilization and workload efficiency are taking center stage. Large AI clusters can contain thousands of powerful servers equipped with tens of thousands of AI accelerators. AI workloads are supercharged by directly connecting compute nodes, and the accelerators within them, to distribute the tasks and increase productivity. As workloads are divided among the accelerators, SiT5977 is critical for synchronizing compute nodes and optimizing efficiency with 3X better synchronization compared to previous generation.
 

Interconnected Clusters Drive AI

Image

Interconnected clusters are at the center of AI infrastructure, containing linked nodes and accelerators required to provide vast and efficient computational power.

Tighter time synchronization also facilitates more accurate telemetry, giving datacenter operators insight to eliminating underperforming compute nodes. In addition, highly tuned synchronization allows architects to reduce the guard band between data packets, known as the window of uncertainty. A smaller window of uncertainty reduces the risk of packets with overlapping time, in turn lowering the frequency of data re-transmits, therefore making the datacenter more efficient.
 

Best Dynamic Stability

Image

SiT5977 offers the best-in-class dynamic stability with ±1 ppb/°C frequency over temperature slope (dF/dT).

The ability to maintain tight synchronization in harsh datacenter environments is essential. However, heat generated by powerful AI processing units combined with cooling fans can create rapid and extreme temperature transients that affect quartz devices, leading to dropped packets and poor synchronization. This is no problem for the SiT5977 Super-TCXO which is resistant to environmental stressors. SiT5977 operates at ±1 ppb/°C frequency over temperature slope (dF/dT) to ensure stability under airflow and thermal shock. This means no performance throttling, smaller windows of uncertainty, and fewer workload restarts, which can cost datacenter operators hundreds of thousands of dollars.

800G Network Connectivity

Equally important to synchronization is fully utilized, high-speed network bandwidth. 800G is becoming essential to support the intensive bandwidth and low latency needs of AI workloads. To support high-speed connectivity, SiT5977 integrates a low-noise PLL, which produces a clean clock edge and exceptionally low jitter, lowering bit error rates. SiT5977 is the first TCXO to offer high-speed differential output for high-speed data communication.
 

Ultra-Low Phase Jitter

Image

SiT5977 offers ultra-low phase jitter at 80 fs from 12kHz to 20MHz to support 800G data.

Low jitter performance is vital to enabling higher bandwidths in next-generation AI datacenters, reducing accelerator idling time while they wait for data. The class-leading jitter performance of SiT5977 removes data bottlenecks and ensures that accelerators run as efficiently as possible. Together, precision synchronization and higher bandwidth utilization increase the productivity of the entire datacenter.

4X Smaller Footprint

By integrating ultra-low phase jitter and high-resolution digital frequency control into a 5.0 mm x 3.5 mm package (compatible with a 5.0 mm x 3.2 mm footprint), the jitter cleaner typically used in the traditional quartz-based timing architecture can be removed. By eliminating the jitter cleaner and using a small TCXO as opposed to a larger OCXO, the timing footprint is made 4x smaller. This allows for the use of larger processors in space-constrained systems, enabling system architects to fully leverage the high-speed 800G bandwidth.

New Architectural Options for Faster Processing

Image

In a Smart NIC application, the small SiT5977 Super-TCXO with its ultra-low jitter and digital output frequency control can eliminate the need for a jitter cleaner. This affords additional real estate for a more powerful processor and enables new architectural options.