Bravo
If ARM was an arm, BRN would be its biceps💪!
Energy is Everything for Edge Computing
By Brandon Lucia, CEO & Co-Founder, Efficient 11.28.2024 0The need for intelligence in the physical world is pushing more sophisticated computation into edge devices. These devices, which previously were simple conduits between a sensor and the cloud, now support complex AI and ML, digital signal processing, data analytics, radio frequency (RF) data processing, and a host of other use cases.
Pushing intelligence into these devices increases the energy demands of these devices, which are already energy-starved in their widespread and far-reaching deployments. The critical need for energy-efficiency requires fundamentally rethinking the computing hardware that we use to build these devices.
At the heart of most edge computing devices, you will find the computing hardware of yesterday—CPUs and FPGAs—which are inefficient and inflexible. Many devices rely on traditional “von Neumann” processors, which waste as much as 90-99% of the energy that they consume due to architectural inefficiencies, leading to needless data movement and instruction control overheads.
While virtuously programmable, traditional “von Neumann” CPUs are just too inefficient. FPGAs are often the first to replace a CPU at the edge, offering a path away from some of the overheads of the CPU and, in some cases, providing improved performance and efficiency. FPGAs, however, are a challenging target for application developers, requiring the specialized skills of a digital design team and a much longer time to market. Moreover, FPGAs were originally designed for circuit simulation and are overspecialized for inessential features, yet under-provisioned for programmability and efficiency. FPGAs do not have a software story, nor do they offer a clear path forward for edge deployments.
On the other hand, some devices are migrating toward GPUs and even more specialized accelerators, which often promise to make an application faster and more efficient. Both require working with new languages and APIs, and as soon as an application does not fit the paradigm, the benefits begin to degrade.
Highly specialized accelerators are also a risky choice, leaving developers with the question: “Will the program that I care about today be the one I care about tomorrow?” If not, designers must discard the accelerator entirely and completely re-design their application around new hardware. On top of this, an accelerator that supports only a narrow strip of an application’s underlying functions (e.g., convolutional neural networks) leaves the remainder of the application unaided and inefficient. Specialization presents a foundational risk to building robust and adaptable edge computing applications.
By shifting to highly energy-efficient, yet general-purpose processor architectures, we can avoid the overheads of von Neumann processors by spatially mapping a computation’s instructions across an array of hardware resources. If spatial dataflow architecture is incorporated, the result of one operation can be directly routed to the input of another operation, according to program dataflow without accessing any intermediate memory.
Additionally, spatial mapping can also minimize the costly data movement in a chip, as it eliminates the price of instruction supply and dataflow. The key to this generality is the co-design of a compiler and software stack to support developers with highly efficient dataflow hardware. The result is a new category of general-purpose processors that are programmable using traditional software, which avoids over-specialization or complicated language while providing orders of magnitude better energy efficiency than leading CPUs.
Especially as edge computing solutions become more integrated with multi-sensor systems, AI, and a broad array of computational demands, the industry needs vastly more programmable energy-efficient processors to alleviate energy constraints across five core industries: smart cities, agriculture, energy and gas, space and defense, and health-tech and wearables.
Smart cities and public sector
Industrial edge devices are used to optimize traffic flow, monitor the health and condition of infrastructure, such as bridges, roads, and buildings, and improve public services in smart cities.However, energy constraints can limit the deployment, density, and coverage of these devices, especially when paired with the cost and effort required to regularly and manually deploy, monitor, and replace batteries. More energy-efficient sensors would reduce the frequency of battery changes, eliminate the need for wired power connections in smart cities, and could be leveraged for continuous infrastructure and public space monitoring, traffic management, pest detection, waste management, smart lighting, and more.
Agriculture
Similar devices are used extensively in agriculture for precision farming, monitoring crop health, agricultural fleet management, and managing resources like water and fertilizer. However, deploying these devices over geographically distributed areas with minimal power sources is also challenging due to the need for frequent battery replacements or recharging.By eliminating the need for battery-related maintenance, farmers can significantly expand their sensor networks for enhanced monitoring and management of crops and operations. This shift would enable more efficient water and fertilizer deployment, leading to improved harvesting practices and pest mitigation, ultimately boosting crop yields and operations.
Energy and gas
Edge devices are also used for real-time monitoring, maintenance, and control of crucial pipelines or power systems, enabling predictive maintenance and reducing downtime. However, the energy constraints of these devices limit the scale of their deployment.In critical infrastructure, where continuous smart monitoring is a requirement, the operational cost of battery maintenance makes large-scale deployments infeasible. By advancing energy-efficient computing in these sensing devices, widespread sensor installations can be enabled—even in remote areas where renewable power sources like wind and solar are prevalent. This not only reduces maintenance time and outages, but also improves public safety, drives down route-based maintenance costs, and fosters more sustainable operations.
Space and defense
In the space industry, edge devices face unique challenges related to energy usage constraints. These devices operate in harsh environments with extreme temperatures, radiation, and vacuum conditions, which can affect their performance and efficiency. Additionally, space missions often rely on limited power sources, such as solar panels or batteries, restricting the energy available for these devices.This limitation is critical as missions can last from months to years, requiring edge devices to operate efficiently without the possibility of recharging or replacing batteries. Communication constraints further complicate energy-saving strategies and optimization efforts, as remote management and updates are limited. Given the high cost of deployment and the limited resources for maintenance or repairs once in space, ensuring the reliability and energy efficiency of these devices is paramount. Ultra energy-efficient processors would open opportunities for increased device lifespans, improved reliability, and more complex on-device operations for data gathering, communications, monitoring, and more.
Health-tech and wearables
Most wearables like smartwatches or smart rings are limited to utilizing small batteries to keep the device lightweight and compact. This inherently limits the amount of energy available for the continuous processing, data transmission and communication tasks these devices are used for. A more energy-efficient processor for wearable devices would not only allow users to go longer in between charges, but greatly improve performance while consuming vastly less energy for the same or even more complex on-device tasks than what is currently on the market.Today, devices spend a majority of their energy channeling data back to a nearby smartphone, offloading AI functionality to the phone and squandering energy on communication. However, new, more energy efficient computer architectures make it possible to perform sophisticated signal processing, analytics, machine learning and even generative AI functionality directly on even the tiniest devices.
Efficient computing locally uses vastly less energy and enables more sophisticated processing for more data collected by the device. Devices will spend the “dividends” of energy efficiency by adding more functionality to smart wearables. This will augment situational awareness, provide real-time translation, and interpret environmental and bio-sensory data to better understand behavioral and lifestyle factors surrounding health and wellness.
As the world continues to shift towards AI-specialized hardware and processors, older or less general-purpose devices are rapidly becoming obsolete, requiring more frequent replacements. This ongoing cycle of hardware replacement causes enormous production costs in both energy and carbon emissions, straining resources and exacerbating environmental degradation.
As more processing, analytics and AI find their way into sensor-enabled devices deployed to the extreme edge, the energy cost of computing becomes a more urgent, existential concern for these critically important application use cases. Addressing this challenge is vital not only for the efficiency and longevity of these devices but also for the sustainability of their deployment in our rapidly evolving technological landscape.
what ever dude… if you can not handle feedbacks and getting immediately impolite “rubbish” then simply don’t answer to my postings … very simple!
