Intel and Broadcom have both suffered major mobile setbacks in the past year. Intel finally had to admit defeat in its serial attempts to become a powerhouse in handsets. Having abandoned its mobile processor and system-on-chip efforts in 2016, Intel then bowed out of the handset modem segment too, when Apple rejected its 5G chip and made peace with Qualcomm (and subsequently acquired those Intel modem assets). Broadcom, meanwhile, had bought and then shut down Renesas’s modem activities in 2013-14, and then tried to acquire Qualcomm, a deal which was blocked by the US government earlier this year.
But failure in handsets does not mean these two chip giants will not see growth driven by 5G. As 5G networks move to the cloud, the platforms on which they run will require increasingly high performance processors and accelerators, while the transport requirements will also escalate, driving demand for high end switch-chips.
In the past week, Intel announced a new series of FPGAs (field programmable gate arrays) designed to support new connectivity protocols and accelerate 5G network workloads in the cloud.
The Intel Stratix 10 DX FPGA is the first to support Intel’s Ultra Path Interconnect (UPI), which claims to reduce latency by 37%; as well as peripheral component interconnect express Gen4 x16; and a new memory controller that supports up to eight Intel Optane persistent memory modules per FPGA. The chip can achieve a theoretical peak transfer rate of 28Gbps.
“The combination of those capabilities as well as our roadmap to the compute express link, or CXL, provides a new level of connectivity, bandwidth and latency capability to our customers,” said Patrick Dorsey, VP of product marketing at Intel’s Network and Custom Logic Group (NCLG).
Intel is targeting the new FPGAs across many markets. “We see a place for it in edge, and in embedded where there’s a real-time low latency capability. We see it in the core network itself, whether that be in NFV or in applications like virtualized RAN. And then in the data center itself,” Dorsey said.
Over at Broadcom, the company unveiled a chip for the physical layer (PHY) of the Ethernet port, supporting dual-400Gig Ethernet and the MACSec security technology (a standard for Ethernet traffic which executes encryption in the PHY layer). New levels of security and reliability are promised by incorporating this device into switches, at a time when Intel has raised its game as a challenger to Broadcom’s switch-chip dominance, with the acquisition of bare metal switch pioneer Barefoot.
The BCM81343 family is designed to support growing bandwidth and security demands in 5G and cloud services. It quadruples the switch bandwidth capability of its predecessor, the dual-100G MacSEC PHY, to 800G in total. It also supports the IEEE 1588 precision time protocol (PTP), to support accurate clock timing for time-sensitive transactions and mission-critical tasks – highly relevant to 5G.
5G services, as well as many emerging cloud applications, increasingly need new levels of reliability, quality of service, real time response and security, hence the focus on hardware-based encryption and accurate time-stamping.
Lorenzo Longo, general manager of Broadcom’s Physical Layer Products division, in a statement. “As cloud and service providers transition their networks to 400G Ethernet to support the growing demand for higher bandwidth and emerging 5G services, it’s imperative that their 400G networks be equipped with 400G MACSec to protect against intrusion attacks, wiretapping, and other threats.”
The new networking device works with 12.8Tbps switches including Broadcom’s Tomahawk 3, Jericho 2 and Trident 4.