Small EdgeQ embarks on Intel-slaying mission in 5G
The silicon implants that power today's 4G and 5G networks tend to come from a small cluster of very big suppliers. Ericsson, Huawei and Nokia are the kit-assembling surgeons in this operation, contributing expertise and stitching parts together. Semiconductor knowhow is provided by the likes of Intel, Marvell and Qualcomm, companies that collectively made almost $130 billion in sales last year. It is a theater occupied by giants with deep pockets, not startups on the lookout for money.
Or was. EdgeQ, a relatively obscure company based in Santa Clara, has barged into the room and been taken seriously by Vodafone, a service provider keen to uncover alternatives to the usual mix of vendors. Founded in 2018, it spies opportunity in open RAN, a Lego-set approach to building mobile networks. Instead of buying a pre-integrated system from one big vendor, an operator using open RAN could pick and mix numerous suppliers – or so the idea goes. EdgeQ literally hopes to slot into the distributed units that handle baseband processing. Trials with Vodafone were announced last month.
Much like Marvell and Qualcomm, EdgeQ is touting an accelerator card that would slide into a server built by Dell, HPE or another such manufacturer. These accelerators are designed to handle the baseband or Layer 1 processing that happens in a radio access network, and to do it more efficiently than any general-purpose chip. But EdgeQ's accelerator has some key differences.
"First of all, we provide a deployable physical layer along with the chip," said Adil Kidwai, the vice president and head of product management for EdgeQ. "Our physical layer is not a reference stack." Essentially, EdgeQ is offering both hardware and all the Layer 1 software that sits on top of it – in contrast to chip rivals focused on the former. "Our business model does not allow us to sell silicon where someone ports their software onto our silicon."
Is this really what operators want? A current bugbear for some telco executives on the technical side seems to be the lack of portability in this market. Today, software designed to run on Intel chips, using Intel's FlexRAN reference stack, could not be used with another silicon maker's hardware, for instance. There is a push for more standardization here to make portability much easier. EdgeQ's tight coupling of hardware and software does not seem to fit easily into this any-silicon-with-any-software narrative.
Yet its deployable physical layer is also highly programmable, according to Kidwai. What this means is that an operator could make changes to algorithms to suit specific needs. "If a carrier comes and says I want to use my own channel estimation algorithm, we are the only company that lets them do that," said Kidwai.
EdgeQ also claims to be ahead of its competitors on power efficiency. Its accelerator consumes 25% to 30% as much power as FlexRAN-based equipment and about 60% to 70% of what other companies would burn, said Kidwai. Operator testimonials are needed before that can be taken as read. But Kidwai attributes his firm's competitiveness partly to its choice of RISC-V – an open-source instruction-set alternative to x86 and Arm, the better-known options used by nearly all EdgeQ's rivals.
"We wrote around 150 custom instructions to do 4G and 5G and AI functionality on RISC-V, but these microkernels are written on bare metal RISC-V processors and so there is no compiler or overhead or anything," he said. Another big attraction of RISC-V as an open-source tool is that its users do not have to pay any licensing fees.
It is not religious about RISC-V, though. Another set of silicon products aimed at the small cell market uses Arm for the Layer 2 and Layer 3 functions in the network. This has prompted some criticism that EdgeQ's overall package is too complex from a software-development perspective. There is worry that its use of customized extended instructions could make software upgrades and the porting of Layer 2 and Layer 3 software very difficult.
But EdgeQ's response is that the customized extended instructions are only applicable for Layer 1 and totally independent of the upper layers. The interface between Layer 1 and those other layers is also compliant with FAPI, an industry standard, the company points out. In the case of a macro cell deployment, Layer 2 and Layer 3 software would run on an external host processor, using FAPI to interact with EdgeQ.
Still, there is bound to be concern that a small company like EdgeQ can outmuscle the Intels and Qualcomms of this world. Today, EdgeQ employs just 200 people across facilities in Santa Clara, San Diego and Bangalore. Together, Intel and Qualcomm had more than 170,000 employees at the end of their respective fiscal years. EdgeQ raised $51 million during a funding round in 2020, and it expects to announce another funding round soon. But Intel and Qualcomm have tens of billions at their disposal for research and development.
What neither one has is a 100% focus on the mobile infrastructure market. "When you think about EdgeQ, this is it and there is no other distraction," said Kidwai. Many of EdgeQ's employees also previously worked for Qualcomm (Kidwai himself came from Intel), giving the company years of combined expertise in this field.
The timing could be auspicious, too. Telcos enthusiastic about open RAN say they are keen to support smaller vendors and extend them a role on network projects. "It is a David versus Goliath story that always excites people," said Kidwai. "We are marching ahead and making waves and offering something people never thought would be offered." A commercial deal with the likes of Vodafone would be a wave of tidal proportions.
- Intel risks losing Arm wrestle as open RAN splits into rival camps
- Vodafone slams Intel and its chip rivals on standardization
- NTT Docomo pitches itself as open RAN pre-integrator
- Vodafone O-RAN guru blesses Nokia, curses SI lock-in and starts build in Germany
- Vodafone aims for massive MIMO in open RAN by March 2023
— Iain Morris, International Editor, Light Reading