With Xilinx in its pocket, AMD gains 5G, Open RAN foothold
AMD will acquire semiconductor maker Xilinx on February 14, 2022, the company announced. AMD said it’s received the necessary regulatory approval, the final hurdle before closing the deal, which was first announced two years ago. In a move to cement its place in High-Performance Computing (HPC) and data centers, AMD first announced plans to acquire Xilinx in October, 2020.
Analysts first valued the deal at $35 billion when it was announced, which sees AMD offering 1.7234 shares of stock for each Xilinx share. But AMD’s stock price has risen steeply since 2020. The final cost to AMD is anticipated now to be closer to $53 billion, making it a tectonic shift in the semiconductor industry.
Consolidation in the semiconductor industry is an ongoing trend, most recently highlighted by Nvidia’s abandoned attempt to buy Arm from owner SoftBank. AMD’s interest in Xilinx comes at a time when semiconductor designers and manufacturers are seeing explosive growth in server product sales.
As more businesses turn to cloud-native operations, data centers need more efficient silicon. That’s why AMD recently reported explosive fourth-quarter growth for its server business. AMD’s server business more than doubled revenue year-over-year because of demand for its EPYC server processors.
But Xilinx’s IP gives AMD a leg up in with its leadership in field-programmable gate array (FPGA) chips. FPGAs use a reconfigurable integrated circuit design, first pioneered in the 1980s by Xilinx co-founder Ross Freeman.
FPGAs sport configurable logic blocks instead of a fixed hardware structure like the conventional CPU and Systems-on-a-Chip (SoCs) designed by AMD. Because of their programmability and flexibility, FPGAs are used to design Application Specific Integrated Circuits (ASICs). FPGAs are known for power efficiency, low latency and high bandwidth.
AMD’s new telecom lynchpin
Besides data centers, military and aerospace, FPGAs are also seen as key to 5G communications and other emerging technologies like vRAN and Open RAN – areas where Xilinx is very active.
Xilinx first introduced it T1 Telco Accelerator Card in September, 2020. The card simplifies Open RAN fronthaul termination at an open distributed unit (O-DU) and Layer 1 offload, according to the company. The T1 card cuts down the number of CPU cores needed to operate an O-DU and, based on the offload of channel coding from CPUs to the T1 card, delivers a 45-x encoding throughput improvement compared to a server running the same functions without the acceleration capabilities. The card also reduces capex and power consumption, according to Xilinx.
Last April, Mavenir announced it was working with Xilinx on a 4G/5G O-RAN massive MIMO portfolio with a 64T/64R configuration that supports up to a 400 megahertz bandwidth.
And in July, Xilinx, Keysight Technologies and Cisco announced a partnership to develop fronthaul solutions to ease CSP migration from 4G LTE to 5G open radio access networks.
Besides FPGAs, Xilinx also makes programmable SoCs, including the Adaptive Compute Acceleration Platform (ACAP). Xilinx describes the architecture as “a hybrid compute platform that tightly integrates traditional FPGA programmable fabric, software programmable processors and software programmable accelerator engines.”
Xilinx has developed ACAP processors for specialized cloud, network and edge applications.
