AI computing in space with NoLoad® NVMe Computational Storage
- Written by: Andrew Maier
Our partner Avalanche Technologies is showcasing the NoLoad® NVMe Computational Storage Processor (CSP) as an interface for the SpaceVPX platform at the SmallSat Symposium in Mountain View, CA, in conjunction with Trusted Semiconductors, Blue Halo, and Mercury Systems. The Space Drive combines the performance benefits of Eideticom’s NoLoad® Computational Storage Platform with Avalanche’s flagship Space Grade MRAM offering, STT-MRAM and specifically targets AI computing in space. Eideticom’s NoLoad platform is an innovative standards-based NVMe CSP solution to support increasing demand from the defense and commercial space communities for computation and storage. With its virtually infinite endurance, radiation resilience and compelling densities, Avalanche’s STT-MRAM provides a robust low latency storage technology for the rigors of space.
Enabling NVMe Computational Storage in Space
As the complexities of space operations grow and the vast distances limit our ability for data transmission back to earth the demand for processing and thus storage in space is ever increasing. Computing in the space industry presents extra challenges compared to terrestrial deployments, requiring unique and efficient solutions. While radiation tolerance is a crucial focus for many solutions, space applications also need to be aware of power consumption, heat dissipation, physical weight, and the added costs of deployment and redundancies. Data transfers are inherently power hungry even in earth-based applications.
“We observe that data movement between the main memory and conventional computation units is a major contributor to the total system energy consumption in consumer devices. On average, data movement accounts for 62.7% of the total energy consumed by Google consumer workloads.” [1]
By moving compute functions closer to the physical storage, we can eliminate the data transfers and thus their required energy use. This is true in both terrestrial and extra-terrestrial applications. Space is the ultimate edge network and computational storage is the ideal technology due to its ability to move compute operations close to storage, minimizing power consumption due to internal data transfers, reducing bandwidth to earth by allowing acceleration of on-board decision making.
NoLoad Computational Storage
Eideticom’s NoLoad platform is purpose built for acceleration of storage and compute-intensive workloads providing transparent computational offload. As a first benefit it provides a performant NVMe compliant interface which enables the use of well-proven standard-based drivers facilitating integration into mission critical space systems. A second benefit is the ability to host computation behind NoLoad’s NVMe interface. Compute resources are identified and accessed as NVME computational namespaces. The combination of NVMe compute and storage enables compute functions to moved closer to storage minimizing data transfers, reducing latency and increasing performance.
Eideticom provides a complete host software and compute stack including application software libraries and computational accelerators such as compression, encryption and AI processing. In addition, device management functions including NVMe Firmware update and the NVMe Management Interface (MI) are supported.
Avalanche STT-MRAM
Avalanche Technology Inc. is the leader in next generation Perpendicular STT-MRAM technology, accepted as the front-runner to replace traditional Flash and SRAM for unified memory architectures in future SOC systems, delivering high performance and low power at 55, 40, 28 and 22nm with scalability beyond 14nm. With a proven STT-MRAM portfolio at multiple geometry nodes combined with an intellectual property portfolio of over 300 patents and applications, Avalanche Technology is delivering on the promise of enabling the next generation of scalable unified memory architecture for industrial, IoT, aerospace and storage applications, making it the true “Next Generation MRAM Company.” For more information, visit us online at https://www.avalanche-technology.com.
Merging NVMe Computational Storage with MRAM
The Noload Space Drive combines Avalanche MRAM with NoLoad CSP technology to enables near-data processing for storage centric, analytic and AI processing workloads. Target applications include storage services i.e. compression & encryption, data format conversion, filtering, metadata management, database analytics, video processing, and AI Inference for tagging & classification.
The integration of NVMe computational storage with Avalanche Technology STT-MRAM, merges the benefits of NVMe with space-validated MRAM storage devices providing efficient and accelerated access to radiation tolerant non-volatile data. The NVMe standard is a low latency high throughput open-source protocol that has widespread acceptance/familiarity via lightweight drivers that exist in all major operating systems. Companies like us (Eideticom) leverage the NVMe ecosystem to enable efficient access to computational storage services, such as data compression, encryption and AI processing. By enabling computational storage services on MRAM devices, we can observe more efficient power usage (operations per watt), reduce deployment costs, and offload critical on-board decision making.
AI Applications in Space and the NoLoad Space Drive
The demand for AI processing in space has witnessed a significant surge, driven by the intricacies associated with transmitting data back to Earth-based servers. Complexities such as power consumption, difficulties with line of sight, and the vast distances involved, especially for missions to destinations like the moon and Mars, underscore the need for efficient onboard processing. Processing this data efficiently in both bandwidth and power usage is critical. Image recognition, object detection, and decision-making processes are crucial components of AI applications in space, with implications for military, civilian, and research-oriented missions. The NoLoad Space Drive can be used to accelerator AI processing applications using Eideticom’s software stack and NVMe based accelerators. A recent whitepaper from our partners, Avalanche Technology, goes into a deeper dive in the future of AI computing in space [2].The combination of NVMe based acceleration with STT-MRAM technology can introduce a radiation tolerant approach to providing efficient AI processing in space applications.
References
[1] A. a. G. S. a. K. Y. a. A. R. a. S. E. a. T. R. a. K. D. a. K. A. a. K. A. a. R. P. a. M. O. Boroumand, "Google Workloads for Consumer Devices: Mitigating Data Movement Bottlenecks," [Online]. Available: https://dl.acm.org/doi/10.1145/3296957.3173177.
[2] A. T. Inc, "AI Computing In Space," [Online]. Available: https://go.avalanche-technology.com/ai-computing-in-space-whitepaper?s=li.