Living on the Edge: A Look at Leading Chipmakers and Their Push for Intelligence
Three dominant megacorporations are empowering edge devices by creating accurate, secure, and low powered chips with capabilities of giant data centers at your fingertips.
Diminishing Machine Learning (ML) Workloads
Arm is a UK-based semiconductor design company that is a prominent manufacturer in the world of AI applications, creating powerful, energy-efficient ML processors to minimize the overall workload for various AI technologies.
The Ethos-U55 NPU
The Ethos-U55 Neural Processing Unit (NPU), developed by Arm, is claimed to be the industry’s first microNPU that delivers fast response time, greater consumer privacy, and longer battery life. Its goal is to increase the system’s inference performance and reduce delivered power for the entire edge device.
By providing a 480x increase in ML performance, it has become a competitive solution for various AI-based IoT applications. The Ethos-U55 NPU is improving object classification, face detection, imaging, and speech recognition for AI technology.
Another unique AI innovation is a socially interactive robot called ElliQ. ElliQ is designed to alleviate loneliness amongst the elderly population. With the traumatic times of COVID-19, a solution to fight against loneliness is needed now more than ever.
Result of adding Ethos-U55 microNPU on typical workloads during voice assisting applications. Image used courtesy of Arm.
Transforming Sensing into Perceiving
The type of security necessary for processors in AI-based applications is crucial and difficult to achieve. Standard devices are relying on giant networks that have high-security threats due to data flow being channeled through the cloud.
A California-based megacorporation, Perceive, is taking a stand on providing the capabilities that are only possible through the cloud at edge device itself.
Unlike the current market that is designing for external cloud usage, Perceive has created Ergo, which provides the functionalities of a large datacenter in the secure location of the edge device. It delivers highly accurate, lower power consumption that not only outperforms other ML processors but does not compromise the security of consumer devices.
The standard 2-5 Tera Operations per Second (TOPS)/Watt, which is used to measure the processor’s performance efficiency, was surpassed by Precieve’s Ergo by an impressive 55 TOPS/Watt, complimented with 20x efficiency and small 7x7mm packaging, Ergo is an appealing option for smartphone and camera integration.
Perceive's Advanced Processor Ergo
Recently, Forbes discussed how life will be easier thanks to the advanced processor, Ergo. Perceive’s Chief Executive Officer, Steve Teig, claims the significance of there edge inference processor and their push for giving consumers the capabilities of a large neural data center at the edge device.
How much of an impact could be made with processors such as Ergo? Well, think about the average microwave and how it requires human interaction to operate.
In the near future, microwaves with edge inference processing will allow the consumer to place a cold meal in the microwave and receive an accurately heated meal without a single push of a button. Embedded devices will not only sense objects but perceive them.
A product image of the Perceive Ergo chip. Image used courtesy of Perceive.
Bringing Cloud Intelligence to Embedded Devices
For decades chip manufacturers have been using the limiting rule-based computer architecture, Von Neumann architecture, to create edge devices with advancing inference processors. This design method only allows processors to have one shared memory and one centralized data manager. These limits result in heavy data traffic, high heat dissipation, and overall insufficient processing.
In order to make advancements, developers need to push the envelope by redesigning the structure at its core. A top-performing AI chip manufacturer, Hailo, has reinvented its process by delivering a newly structured computer architecture that goes beyond handling sensors and remote data streaming.
The Hailo-8 Processor
The Hailo-8 processor brings cloud intelligence to embedded devices with breakthrough specialized deep learning capabilities. How did Hailo become a gamechanger? Here is how; while most edge inference processors are using external storage or single centralized form of management, Hailo-8 gives each unit dedicated storage and controlled management.
Hailo-8’s redesign is more economical by operating in real-time with minimal power consumption with a performance efficiency of 2.9 TOPS/Watt. Hailo has received recognition by becoming a CES 2020 Innovation Award Honoree due to its brilliant design. Halio aims to empower the device without compromising consumer security thus putting them at the front foot of a new era of chips!
A New Era of Intelligence
By comparison, each edge inferencing processor focuses on improving our daily living. Perceive and Arm deliver solutions for consumer used devices while Halio aims to create smarter cities and predictive maintenance for industrial manufacturing.