A pioneering new approach to embedded systems that would allow this kind of interaction was recently launched by a startup co-founded by Brandon Lucia. The innovation—called E1—takes advantage of an unconventional architecture that greatly increases the efficiency of their processing. It promises 10x to 100x greater performance compared to today’s ultralow-power CPUs for common embedded applications.
The E1 is based on a unique architecture that breaks with the conventional von Neumann paradigm. The E1 departs from this model radical. This approach employs a unique mesh of “tiles” to tightly couple both processing and memory. This approach lays out a program’s commands right on the chip. It powers high-level functions such as fast Fourier transforms on sensor information as well as convolution functions, both of which are fundamental to machine learning.
The E1’s functionality primarily focuses on the effcc Compiler. It regularly crunches through projects developed in C and other widely used languages. This specialized compiler is responsible for distributing each instruction in a given program to a particular tile within the E1’s array to ensure fast execution. Lucia emphasized the performance capacity of the E1, stating, “Several billion times per second, you’re pulling an instruction in from memory. That operation costs some energy.” This makes the energy efficiency feature front and center, while keeping the high-speed processing abilities.
Rakesh Kumar, a computer architect at the University of Illinois Urbana-Champaign, said the implications of Lucia’s work are profound. He explained that ultralow power firms have struggled with strong price collapse from very competitive competition through the low-power microcontrollers marketplace. Ultralow power firms have suffered due to strong competition in low-power, super cheap microcontrollers. The real challenge is in defining a new capability,” Kumar said. The E1 seems to overcome this hurdle by providing efficiencies not achieved with earlier architectures.
The E1’s architecture supports its rich fabric, unlike other architectures its fabric can move values over feedback paths which is necessary for general purpose computing. This unique flexibility is one key feature that differentiates it from other dataflow-style architectures. Those legacy systems limit the data channels to satisfy the needs of software. Lucia compared the E1’s operational mechanics to a railroad switch track. This analogy underscores its remarkable sophistication to quickly and effectively reroute commands as dictated by computational requirements.
By taking advantage of this groundbreaking architecture, the E1 promises to transform the performance embedded systems deliver on the projects they touch. It tends to perform best in use cases that require very fast inference and low power requirements. This combination makes it a robust player in a crowded marketplace. That kind of efficiency could set a new bar for embedded systems. It will move the market from simple to more sophisticated and capable solutions.