Chip startup Efficient Computer Corp. recently exited stealth mode, with a $16 million seed funding round led by Eclipse, and announced a transformative computer architecture and software stack to deliver what it calls unprecedented energy efficiency for general-purpose edge computing. The new computer processor is up to 100× more energy-efficient than leading general-purpose CPUs, while delivering 1,000× lower power than today’s GPUs, thanks to the company’s reconfigurable dataflow processor Fabric architecture.
Artificial intelligence requires constantly deployed intelligence on an enormous scale. Until now, challenges to achieving that pervasive intelligence included communicating vast amounts of raw data from the edge over rapidly draining device batteries and existing processors that make local computation too inefficient.
Efficient’s processors can run for multiple years on battery power. Pre-processing and compression of data occurs before transmission without impacting energy efficiency, and local computation takes place on-device for greater efficiency.
Energy is key in system architectures, operating costs and carbon emissions. To address energy consumption, companies today are designing specialized “accelerator” architectures that improve efficiency, but only on a fraction of the applications. Efficient Computer was founded to address this dilemma and create the most energy-efficient general-purpose processors possible.
“Energy consumption impacts nearly everything in modern computing, from where devices are located to the capabilities they offer and the scale of their deployment,” said Brandon Lucia, Efficient’s co-founder and CEO. “We are removing the energy barrier from computing at the edge while giving developers the freedom and flexibility to quickly build devices and applications at scale. Efficient hardware and software will significantly reduce energy consumption for computing, creating entirely new categories of use cases.”
Lucia explained: “CPUs and microcontrollers suffer from von Neumann’s inherent energy inefficiency and waste 99% of energy. FPGAs present an alternative that, in some cases, can be more efficient than CPUs. However, programming FPGAs is difficult and costly, requiring a hardware development flow that is unfamiliar, cumbersome and unproductive for most developers.
“GPUs can be highly energy-efficient, but only on so-called embarrassingly parallel programs, and must be programmed in custom, proprietary languages, such as Nvidia’s CUDA,” he added. “In contrast, Efficient’s Fabric architecture is programmable using familiar, standard programming languages with orders-of-magnitude better energy efficiency.”
According to Lucia, the company’s new architecture provides three key benefits that sets it apart from other computer chips. It makes computing orders-of-magnitude more efficient, prolonging battery life and increasing computing capabilities. It is developer-friendly and future-proof, supporting mainstream, high-level programming languages.
Efficient, therefore, frees developers from learning cumbersome new languages and tools that come with specialized hardware, Lucia said, and finally, the processor allows for more efficient on-device computing solutions and computing capabilities at the edge.
The low-energy-consumption technology in edge devices also avoids costly and often unavailable communication to the cloud, and it improves efficiency and data privacy, he said.
The Fabric architecture approach
Efficient’s Fabric architecture took a decade of research at Carnegie Mellon University, developing the building blocks to make computing up to 100× more efficient. The approach spans hardware and software.
Caption: Monza test SoC (Source: Efficient Computer Corp.)
The Fabric architecture expresses programs as a “circuit” of instructions that explicitly shows how instructions talk to each other. This allowed laying out the circuit spatially across an array of extremely simple processors and to execute the program in parallel, with much simpler hardware and less energy than any existing processor, according to the company.
The solution enables new classes of products that combine sophisticated computational features and longer battery lifetimes. General-purpose software development flow allows rapid iteration of designs at orders-of-magnitude lower cost than FPGA or custom silicon hardware development.
“Efficient has created a transformative computer architecture and software stack that delivers unprecedented energy efficiency for general-purpose computing,” Lucia said. “Efficient’s novel Fabric architecture is a unique ‘post-von-Neumann’ approach to general-purpose computation that eliminates the extreme energy overheads of existing general-purpose processor architectures.”
As a result, it enables dramatically more energy-efficient chips, bringing computing capabilities to use cases that were previously impossible due to the limited energy availability and the high energy cost of computing, he added.
Examples cited include a smartwatch or health wearable that could go weeks between charges and support more sophisticated sensor data analytics, or a remote-sensing satellite that could provide high-fidelity data insights using the limited energy of a nanosatellite’s small battery. Also, an infrastructure monitor could be deployed with a decade-scale lifetime, running sophisticated AI and data processing.
The Fabric processor architecture is implemented for the first time in Efficient’s Monza test system-on-chip (SoC). It is in the lab for “bring-up” and measurement, Lucia said. “The Fabric’s compiler was designed alongside the hardware from day one, and it compiles programs written in high-level C, C++ and other high-level languages and frameworks.”
The Fabric architecture, implemented in the Monza test SoC, can provide up to 100× better energy efficiency than industry-leading embedded von Neumann processors, and unlike specialized hardware, it can be programmed in software using a standard, familiar development flow, he said.
“In terms of differentiation from other solutions to edge processing, the Fabric architecture’s energy efficiency enables more sophisticated computation, without going to the cloud,” Lucia said. “The Fabric architecture allows computing on-device for 10× to 100× less energy, compared with other solutions to edge processing.”
The Efficient processor saves energy across applications like machine-learning-enabled extreme-edge machine vision, continuous audio intelligence, and sensory and signals intelligence. Developers can bring their own code, with support for mainstream embedded languages, to build intelligence applications that integrate AI/ML, signal-processing data analytics and other general-purpose data-processing computations.
Early-stage relationships are under development with customer partners to create proof-of-concept devices that demonstrate extreme energy efficiency. Those in its early-adopter program will have access to the Efficient software development kit and hardware.
Advertisement
