EdgeQ to Debut Single Chip for 5G Private Networks

//php echo do_shortcode(‘[responsivevoice_button voice=”US English Male” buttontext=”Listen to Post”]’) ?>

EdgeQ, a startup led by former Qualcomm executives, plans to introduce its first chip for 5G private networks in the middle of this year. Before that, the company is set to hold a technology demonstration with customer Vodafone at the Mobile World Conference (MWC) in Barcelona near the end of February.

EdgeQ CEO Vinay Ravuri sat down for an interview with EE Times about the company’s plans to kickstart the 5G private network business. Some say the “Q” in EdgeQ stands for Qualcomm.

“Some people do joke, ‘It’s Qualcomm with an edge,’ and it’s Qualcomm this and that,” Ravuri said. “But no, it’s not meant to be any of those. The genesis of the company is to merge compute and connectivity together.”

EdgeQ, which will make its first commercial chip early this year with support from fab partner Taiwan Semiconductor Manufacturing Co. (TSMC), expects to earn its first revenue a few months later. This month, the company plans to demonstrate how its solution works with global telecommunications provider Vodafone and ecosystem partners at MWC 2023.

“Vodafone is interested in working with EdgeQ on the open RAN (Radio Access Network) aspect,” Ravuri said. “Vodafone is interested in this because there is a huge power benefit.”

Open RAN, or O-RAN, is the final link near the antenna in a mobile network that has been opened to let operators integrate equipment from multiple vendors. Two of the EdgeQ chips on a card will replace a rack of servers in an O-RAN installation, Ravuri said.

“Often, these things sit outside in a ruggedized environment, and the space and power that is provided to that system is going to be limited. You can’t put a big shelf of a server rack under a tower. Space is a precious commodity, and we make a big dent there. We are one-tenth the space because an entire box is replaced with one card.”

EdgeQ is targeting two markets: macro cells for telcos like Vodafone and small cells for operators of 5G networks in warehouses, ports, farms and even defense installations.

No wide adoption yet

To operate a private network, finding a free cellular spectrum is the first hurdle, Bob O’Donnell, president of TECHnalysis Research, told EE Times. Most of the spectra are licensed, controlled and paid for in billions of dollars by carriers like Vodafone.

There are situations where the carriers can sublicense some of that spectrum, according to O’Donnell. The free-spectrum alternative that people are pushing hard for is Citizen’s Broadband Radio Service (CBRS), he added.

“CBRS has not had wide adoption just yet, and that’s a huge issue,” he said. “Because it’s unlicensed, anybody can use it, meaning two buildings right next to each other could theoretically be used in the same CBRS spectrum.”

There are still more problems with CBRS, he added. Users may buy preferred access to CBRS, but if they are near a seacoast in the U.S., the military still has the right to take over those frequencies.

EdgeQ says that its solution supports a range of spectra, from sub-6 to mmWave, as well as CBRS.

From left to right: Feng Kan, head of platform engineering; Tae Ryun Chang, sr. director of hardware; Chaekwan Lee, head of PHY; and Vinay Ravuri, CEO and founder. (Source: EdgeQ)

Software story

Another big issue for 5G private networks is setting up and maintaining the cellular infrastructure, according to O’Donnell. “No one has really nailed that yet. That’s more of a software story.”

That may be how EdgeQ stands apart.

Cellular is so difficult that only a handful of rivals like Nokia and Ericsson are in the business, Ravuri said.

“You have to have chip-design people, you have to have people with an information theory background. You have to have people that have an algorithmic-implementation background, and you have to have people who have seen this in practice because this is not all science. There’s some art here.”

The EdgeQ chip has a composable unit that can be upgraded over the air using software.

“For me, that is the key differentiator that is soft and for different standards of 5G,” Ravuri said. “Different applications can get different firmware rather than having hard-coded chips. That’s the big thing because I can be very flexible.”

EdgeQ says its “base station on a chip” should be as configurable and easy to deploy as WiFi systems are today.

“There is not a chip today that covers the gamut all the way from Layer 1 to Layer 3 [of the Open Systems Interconnect model],” the company said. “We take in a bunch of different technologies—4G, 5G: We’ve taken AI into it because computers are part and parcel of this entire thing. We’ve taken timing synchronization, we’ve taken different types of MIMO [multiple input, multiple output] technology and put it into a single chip. We’re the only one in the industry.”

The EdgeQ chip allows customers to customize their own PHY layer as part of the system if necessary. In the seven-layer OSI model of computer networking, the PHY, aka layer 1, is the lowest layer most closely associated with the physical connection that provides an electrical, mechanical, and procedural interface to the transmission medium.

Ravuri offered the example of a defense customer that wants a secure system: “They don’t want me to know about it [the PHY] because it’s national security. They develop that technology, and they want to change it and keep it to themselves. [Defense is the] perfect application for that.”

5G has many more applications, Ravuri added.

A company like Honeywell might want a 5G system for a factory making engines and electronics for airplanes, he suggested. Much of the solution is downloadable.

“A single chip is just different software products for me,” he said.

Nokia and Ericsson offer products for 5G networks that EdgeQ will compete with.

“We want to enable everyone else,” Ravuri said. “Someone who can just make a box like WNC [Wistron NeWeb Corp. of Taiwan]. They are very good at making a box of hardware, but they’re not very good at the cellular technology.”

To add users or throughput to an EdgeQ-based system, there’s downloadable software and firmware, according to the company. One download provides ultra-low latency. For backward compatibility, to 4G or even WiFi, there’s another download.

Beating benchmarks

EdgeQ created a chip architecture to support compute and connectivity in parallel. The same 60 or so RISC-V cores that can handle 5G baseband processing on the chip can also be refactored to handle both 5G and AI.

“Compute comes in two flavors for us,” Ravuri said. “One is machine learning. A camera in the system that wants to make intelligent decisions runs the models on RISC-V CPUs that are capable of doing both 5G and machine learning simultaneously in a single place. On top of it is an ARM CPU complex which acts much like the cloud in terms of the elasticity, meaning it is virtualized, it can do virtual machines. It can do all the things that the cloud does, but at a lower scale.”

It’s still early days for 5G private networks, but any use case that involves the generation of large amounts of data will benefit from putting computing resources closer to data creation points, said Will Townsend, a VP with Moor Insights & Strategy.

“Edge can supercharge 5G use cases, reducing latency even further and improving performance and application responsiveness,” he told EE Times. “EdgeQ aims to do this with highly integrated silicon for RAN applications that marries AI and machine learning for improved performance, lower cost and rapid design.”

The bill of materials for a baseband box made with the EdgeQ chips can be half that of the $1000 systems currently offered by Ericsson and Nokia, Ravuri said.

EdgeQ’s power consumption per box is in the range of 13 to 19 watts, half as much as existing products, according to Ravuri. That creates new use cases for the EdgeQ systems, he added.

“Not just remote, even indoor. Inside a building, typically power is not unlimited. You provide it through an Ethernet wire.”

EdgeQ says it is the first vendor to enable power over Ethernet.

TSMC partnership

Ravuri said EdgeQ, despite being a small startup, quickly convinced TSMC to become its foundry partner.

“As soon as they met me, they’re like, okay, you’re not a college kid. We know what you’ve done, and we know your team.”

Ravuri mentions design mistakes that were made in “previous worlds”.

EdgeQ said it uses an older version of TSMC’s FinFET process that outperforms Qualcomm chips made with a sub-10–nm process at TSMC.

“Often what happens in big companies is even though you know there’s a better way, you don’t implement a better way because you just have to maintain what was done,” the company said. “You can’t just clean slate something. Given the opportunity we had to clean slate it, all the baggage went away. That’s how we are able to be in TSMC’s higher-node numbers but achieve better power.”

The CPU technology that EdgeQ has implemented is very innovative, Ravuri said.

“It’s probably the most advanced both for RISC V as well as base stations in general.”

Scroll to Top