The International Solid-State Circuits Conference (ISSCC) is the premier global forum for the presentation of advances in solid-state circuits and systems-on-chip. It was a delight to have been invited to participate in this year’s event as part of the forum on space-based communication and computing, where I spoke about the role of Efficient’s technology as a key enabler for future, SWaP-constrained space-based computer systems.
Fueling the Future of Space Innovation and Exploration
The forum took a broad view on the current state of space computing and communication. In my talk, Enabling the Next Era of Space-Based Cyber-Physical Systems Applications with Ultra-Energy-Efficient, General-Purpose Computing Hardware, I described the exciting challenges in building capable constellations of small satellites for a wide range of applications, including infrastructure monitoring, safety and security, and search and rescue.
Space is hard and in space, every Joule of energy counts. A satellite designer needs to minimize SWaP -- size, weight, and power -- to build a system that is simultaneously capable, while remaining cost effective to launch and operate. Even as they’ve improved in SWaP, satellite-based sensors, like cameras and hyperspectral instruments, capture orders of magnitude more data than a satellite can feasibly communicate to Earth. The consequence of this data excess is that a satellite can operate its sensor continuously or near-continuously, but most of the data are lost because they can’t be downlinked. Recently, the space community has shift toward “orbital edge computing,” which moves data processing computation into the satellites themselves. By computing on device, the satellite avoids the need to communicate with infrastructure on Earth to process the sensor data. On-device computing in a highly SWaP-constrained environment, however, requires careful system design to avoid using excessive energy while processing data, which would in turn require duty-cycling the satellite, limiting its sensor capture rate, or its data processing rate. If the processors in the satellites are inefficient, on-device computing fails to deliver on its potential and mission are forced to an unsatisfying dilemma: lose data not downlinked, or lose data not processed.
Computationally-capable satellite constellations need new efficient processors capable of breaking out of this dilemma and at Efficient, our ultra-efficient, general-purpose E1 processor is the solution to this problem.
Efficient’s E1 Processor brings Energy Efficiency to Space-Based Computing Hardware
In the era of orbital edge computing, efficient computation is critical for extracting value from a satellite deployment. Satellites perform a wide range of computations and require efficient computation, not just for a specific set of functions (e.g., machine learning). A single satellite runs a diversity of algorithms: visual and time-series signal processing algorithms, guidance/navigation/control (GNC) algorithms, state estimation and Kalman filters, computer vision, optimization, machine learning, generative AI, and a wide range of application-specific logic and analytics. It is impossible to anticipate in advance what a satellite’s computational payload may be, which requires efficient hardware for general-purpose computation. Dedicated accelerators for specific functions (like AI chips or even GPUs) constrain developers and do not support the full range of software use cases in a satellite system.
The importance of generality all comes down to Amdahl’s Law, which is one of the most important results in computer architecture: a system is bottlenecked by the part left unoptimized. A case in point is a satellite that incorporates a fixed-function accelerator (even a very efficient one) for some of its functions, but that must rely on a relatively inefficient traditional von Neumann CPU for the remainder of its code. Even if the accelerator (impossibly) reduces the energy cost of the code it accelerates to zero, the remaining portion of the application determines the end to end energy cost. The figure below illustrates the point. Here, about 60% of the application fits the accelerator. The 40% that doesn’t fit the accelerator limits the end-to-end benefit to about 2.5 times, because even if the 60% of the application were to cost nothing, the remaining 40% is unchanged.
Efficient’s Fabric architecture accommodates the diversity of computations that make up modern space systems applications, providing increased efficiency for the entire application. Instead of the 2.5 times best-case end-to-end benefit of the accelerator, for Efficient’s E1 processor the sky (or deep space?) is the limit in terms of its efficiency benefit. E1 supports the entire application, which is the key to huge end-to-end efficiency improvements.
I am extremely excited to see the Efficient E1 processor launch into our lead customers’ use cases later this year, in the space domain, but across several other domains as well. The Fabric brings a level of efficiency for general-purpose computing that just doesn’t exist in the market today. It will be amazing to see application developers leverage Efficient’s architecture to go beyond the limitations of today’s systems and bring revolutionary new capabilities to their customers and use cases. There’s so much to do in space today, from LEO to the moon, and beyond and at Efficient we are leading the way toward SWaP-optimized energy-efficient general-purpose computing solutions for space-based computer systems.
