Artificial Intelligence is creating massive demand for efficient high bandwidth (i) data acquisition from the real-world, (ii) data transportation to the data center, and (iii) data connectivity within the data center.
Omni Design plays a key role in all the following three key segments of the AI infrastruct/data pipeline with it's Wideband Signal Processing™ solutions.
Real-World Data Acquisition
Omni Design’s low-power wideband converters and Wideband Signal Processing™are used in real-world applications like Radar, LiDAR, and spatial computing sensors to digitize real-world data for AI systems.
Data Transport to Data Centers
Omni Design’s data converters an Wideband Signal Processing™ facilitate wireless and wireline data transport, ensuring large-scale data reaches data centers efficiently.
Connectivity within Data Centers
As machine learning models grow in complexity, Omni Design’s solutions Wideband Signal Processing™ support high-speed, low-power data connectivity within data centers, where SerDes IP accounts for about 25% of server power consumption.
SWIFT™ Data Converters:
High-Bandwidth Data Acquisition and Wideband Signal Processing™: SWIFT™ data converters, with sampling rates reaching tens of gigasamples per second (GS/s) and resolutions from 6 to 14 bits, enable the capture and processing of wideband signals crucial for applications like real-world data acquisition for AI (e.g., Radar, LiDAR, spatial computing sensors).
Low Power Consumption: Designed with ultra-low power in mind, the SWIFT™ family contributes to energy efficiency in power-constrained environments, which is critical in dense data center deployments and edge AI devices.
High Linearity and Low Noise: These converters minimize spurious frequency components and distortion, ensuring clean signal conversion essential for high-fidelity Wideband Signal Processing™ required in communication systems within data centers.
Integration with Analog Front-End (AFE) Solutions: Multiple instances of SWIFT™ data converter IP blocks can be integrated into complete AFEs, in addition Omni Designs' capless low dropout regulators (LDOs), precision voltage/current references, and high-speed clock management and PVT monitoring can be incorporated. This simplifies the design and reduces the footprint of complex systems.
Support for High-Speed Communication Standards: SWIFT™ data converters are vital building blocks for enabling high-speed data rates (up to 1.6Tbps and beyond) in fiber optic transceivers and high-speed Ethernet (up to 1.6TbE), which are fundamental for data center networking and high-bandwidth AI training and inference applications.
Direct RF Sampling Capabilities: Certain SWIFT™ ADCs with high sampling rates (multi-GSPS) are suitable for direct RF sampling, simplifying receiver architectures in wireless communication and potentially in advanced sensing modalities for AI.
IQ Data Converter Options: Availability of IQ ADCs and DACs facilitates efficient processing of complex modulated signals used in communication systems within data centers for high-speed data transfer.
Omni Design’s Thermal and Reliability Sensing
The OmniTRUST™ product line including process detectors, voltage monitors, and temperature sensors (PVT) which serve as the fundamental building blocks for monitoring key metrics throughout the silicon lifecycle, from initial design to in-field operation.
In today’s leading machine learning hardware, power and thermal management are critical to optimizing performance and reliability. OmniTRUST™ PVT monitors with accurate temp sensing, glitch detection, and process monitoring, implemented in some of the most advanced processing nodes, allows customer to continuously monitor and react to conditions that could have adverse effects on operation. This can allow precise throttling of performance up to the device limits, when needed, and rapid glitch detection provides feedback about operating conditions whether natural occurring, or due to malicious activity.
OmniTRUST™ PVT (Process, Voltage, and Temperature) Monitoring Solutions:
Enhanced System Integrity and Reliability: OmniTRUST™ PVT monitors provide continuous, localized monitoring of critical silicon parameters (process variation, voltage fluctuations, temperature gradients). This allows for early detection of potential issues that could affect the reliability and longevity of AI and data center hardware.
Dynamic Voltage and Frequency Scaling (DVFS) Optimization: Accurate temperature and voltage sensing enables precise DVFS, allowing systems to dynamically adjust operating frequency and voltage based on real-time conditions. This optimizes performance while minimizing power consumption, which is crucial for data center efficiency and managing the thermal constraints of high-performance AI hardware.
Mitigation of Silicon Aging and Reliability Issues: By continuously monitoring PVT conditions, systems can proactively adjust operation to mitigate long-term silicon aging effects, extending the lifespan and ensuring consistent performance of AI accelerators and data center infrastructure components.
Glitch Detection: High-speed glitch detection capabilities provide rapid feedback about operating conditions, which can be indicative of potential hardware anomalies or even security threats (side-channel attacks).
Process Variation Monitoring: Information on the process variation of core transistors allows for better characterization and optimization of the silicon for specific AI and data center applications.
Simplified Integration: The availability of single core-voltage supply rail PVT monitors simplifies integration into advanced process nodes (e.g., 3nm), reducing design complexity.
Compact Footprint and Low Power: OmniTRUST™ PVT monitors are designed with a small footprint and low power consumption, minimizing their impact on the overall power budget and area of complex SoCs used in AI accelerators and data center equipment.
Programmable Sensor HUB: Some OmniTRUST™ solutions include a programmable sensor hub, offering flexibility in configuring and interfacing with multiple sensors for comprehensive environmental monitoring within AI and data center systems.
