Marvell Technology plans to roll out its new generation of chips based on the world's most advanced 5-nm node by the end of next year, trying to get ahead of rivals in the market for chips used in cloud data centers, corporate networks, 5G telecom systems, and other types of data infrastructure.
The Santa Clara, California-based company said it has partnered with TSMC to develop a new portfolio of chips based on its 5-nm process for a wide range of areas, from data centers to 5G network infrastructure. Marvell said it plans to upgrade its products from TSMC's 7-nm process node, the current state-of-the-art in the global chip business, to the 5-nm node to improve on the storage, bandwidth, speed, and machine-learning capabilities of its chips in a low-power envelope.
Marvell said TSMC's 5-nm technology can improve the performance of its products up to 20% and reduce power usage by up to 40% compared with TSMC's 7-nm node, which was released in 2018.
The Silicon Valley company plans to focus on cloud data centers, enterprise networking, and 5G infrastructure as its core businesses. It is also expanding into the automobile market, which the firm sees as one of the highest growth segments of the semiconductor industry in the long run. Marvell said contracts with TSMC are already in place for many of the chips in its new portfolio. Plans are to start sampling the first 5-nm products by the end of 2021, with mass production soon after.
Marvell is trying to transform itself into a global leader in the infrastructure market with a focus on storage, networking, compute, and security capabilities in everything from data centers to telecom equipment.
The partnership puts Marvell near the front of the line for TSMC's most advanced technology, where production capacity is limited due to rising demand, said Patrick Moorhead, president at Moor Insights and Strategy. "TSMC only partners with a very select few companies on the bleeding-edge process nodes such as 5-nm," he added.
TSMC's largest customers include Apple, Qualcomm, Broadcom, and AMD. Recognized as the world's largest contract chip manufacturer, TSMC makes chips for about 500 customers in all.
Marvell's plan is to move most of its core products for the data infrastructure market to the 5-nm node from TSMC, packing more performance and memory into a smaller die at lower power draw.
In May, the company rolled out its latest Octeon line of network infrastructure chips, which can be used in 4G and 5G networking gear and switches, secure gateways, routers, and other solutions in data centers. They should start shipping to early customers by late 2020. Marvell said Octeon is the world's most popular data processing unit (DPU) for use in data centers and other infrastructure that enables acceleration and offload capabilities, including Smart NICs and security accelerators.
It is also challenging Broadcom in the market for networking-switch chips for areas spanning data centers to autonomous cars. Moreover, Marvell is trying to strengthen its stronghold in switch chips for corporate networks with its latest generation of Prestera ASICs and Alaska Ethernet PHYs, which it revealed last month. Also receiving an upgrade is its Thunder X line of multicore central processing units (CPUs), based on Arm Holdings technology, for use in huge cloud-computing data centers.
Marvell also recently rolled out its new Octeon Fusion family of chips to handle the throughput and latency of 5G networks, which support far faster data transfers than 4G networks. The company is gaining ground in the telecom-equipment market after it partnered with Nokia, which accounts for around 20% of the total market, to roll out a custom-designed chip with Octeon Fusion IP inside. Marvell, which is battling Intel and Broadcom in the 5G market, is also selling chips to Samsung.
"Marvell’s leaning into the leading-edge of process technology also means that the company’s roadmap is aggressive and will likely continue to gain new customers and grow existing relationships well into 2022," Moorhead said
The firm is also expanding its business selling custom chips to large customers. Google, Microsoft, and other titans of the technology industry have started to roll out internally developed chips to run workloads, including artificial intelligence to 5G networking, faster and more efficiently than general-purpose hardware. Marvell is trying to persuade other technology giants, large corporations, and telecom equipment vendors to use its silicon-as-a-service instead of striking out on their own.
Customers can pick and choose any block of technology from its vast IP portfolio, including Arm-based CPUs, networking switches, storage controllers, Ethernet PHYs, high-speed SerDes, and encryption accelerators. Marvell said they can take advantage of any of the leading standard products in its library. The parts are assembled in a way that bolsters the performance and energy efficiency of the final product and reduces the time it takes to bring the IC to market.
"The breadth of IP we're offering to customers is what sets us apart," said Matt Murphy, Marvell's chief executive officer, on a quarterly earnings call last month. "Customers can reuse Marvell's hardened and widely deployed IP and focus their engineering team on only the unique aspects of their application, benefiting from faster time-to-market and lower risk by using our proven technology."
Marvell said that it can also use IP designed by its customers and transplant it in the floorplan of its standard products to create semi-custom designs. It has also started licensing out its standard products, including the Octeon Fusion family of chips, so that customers can add them to their homegrown chips and differentiate themselves in data centers and other areas.
Marvell last month landed a contract to custom-design a chip for a "tier-one hyperscale" customer. "We hope this is the first in a series of design wins with this important customer," Murphy said.
Marvell is bringing its leading portfolio of intellectual property, including processor subsystems, on-chip interconnect fabrics, physical-layer interfaces, encryption engines, chip-to-chip links, high-speed SerDes up to 112 Gb/s, and other building blocks, to the network infrastructure market. These technologies and more are in development right now on TSMC's 5-nm node, the company said.
Marvell indicated that this is a major turning point because it brings the pace of improvement in infrastructure chips closer to the cadence of the smartphone market. Apple recently introduced what it called the world's first 5-nm chip, the A14, for its iPads and iPhones. The chip contains more than 10 billion transistors, a generational leap of 40% compared to the A13 chip, which uses 7-nm.
Raghib Hussain, who leads Marvell's networking and processor business, said the 5-nm technology from TSMC delivers "world-class power, performance, and gate density" and, he added, is "critical for the demands of the world's leading companies in 5G, cloud, enterprise, and automotive."
The company revealed it has also partnered with TSMC to use more advanced nodes, including the 3-nm node. That gives customers in the infrastructure market a long-term roadmap, said Marvell.
The release of each new generation of wireless technology every decade or so has enabled mobile communication to progress considerably since the first portable phones appeared in the 1980s. Technical advances have created new services and business opportunities, driving what’s being referred to as the third wave of communications. The evolution made possible through 5G and future 6G technology will support even more new services for industry and society, well into the 2030s and beyond (Fig. 1).
1. There have now been five generations of mobile communications technology and services, with a sixth on the way.
5G represents the first step toward this next wave of services with expanded connectivity and a significant upgrade in multimedia capabilities combined with artificial intelligence (AI), machine learning (ML), and the Internet of Things (IoT). 5G will be the first generation of mobile communications to utilize millimeter-wave (mmWave) band frequencies, supporting bandwidths of several hundred megahertz and actualizing ultra-high-speed wireless data communications of many gigabits per second.
This article examines the expansion of 5G/6G wireless communications to new areas of service that will drive another industrial wave and offer greater business value for industry and society.
The Third Wave of Wireless Communications
5G and future systems will close the gap between the physical and cyber worlds. Today, mobile consumers use wireless connectivity to access the web from almost any location. In the future, high-speed coverage will be more widespread and faster, with greater emphasis on uplinking information from real-world events, either human and/or IoT, to the internet.
Once this information is in the cloud, AI could reproduce the real world in cyberspace and emulate it beyond physical, economical, and time constraints, so that “future prediction” and “new knowledge” can be discovered and shared. The role of wireless communications in this cyber-physical fusion is assumed to include high-capacity and low-latency transmission of real-world images and sensing information, and feedback to the real world through high-reliability and low-latency control signaling.
Radio communications in this cyber-physical fusion scenario correspond to the role of the nervous system transmitting information between the brain and the body. Communications convert real-world events to the cyber world through enhanced uplink capabilities and provide feedback information to humans and devices through low-latency downlink functionality.
Next-Generation Areas of Service
2. These are some of the new business services that will be enabled with the third wave of communications, which has been initiated with the advent of 5G.
The next wave of communications focuses on three areas of service (Fig. 2) including:
- Enhanced mobile broadband (eMBB), which extends the current mobile experience with high data throughput on the order of more than 10 Gb/s, high system capacity on the order of more than 1000X that of LTE, and a much better spectral efficiency (3-4X) than LTE. Its use cases are high-speed mobile broadband, virtual reality, augmented reality, gaming, and more.
- Ultra-reliable, low-latency communications (URLLC), which focuses on achieving low latency, high reliability, and high availability. The expectation is for latencies of less than 1 ms. This is basically for mission-critical use cases and applications.
- Massive machine-type communications (mMTC), which provides connectivity to a huge number of devices whose traffic profile is typically a small amount of data (spread) sporadically. Consequently, latency and throughput aren’t a big concern. The main concern is the optimal power utilization of those devices because they’re battery-powered and the expectation of battery life is around 10 years or so.
Current activity in mmWave front-end design, including antenna-in-package (AiP) phased arrays (Fig. 3), large-scale beamforming RF integrated circuits (RFICs), multi-technology integration, and system-level electromagnetic (EM) analysis will all contribute to realizing New Radio (NR) access technology that can be cost effective and easy to install. This will support the small-cell networks that achieve 5G/6G performance.
3. The images depict dual-polarized AiP phased-array technology designed in Cadence AWR Design Environment software.
Early 5G deployment and related trials have shown room for improvement in coverage and uplink performance in non-line-of-sight (NLOS) environments and heavy traffic use cases. While future system performance goals are still in the early phases of consideration, full realization of the promise of this next wave of communications creates a need for continued enhancements. Extreme performance is necessary to provide the high reliability and low latency called for by mission-critical (time-sensitive) applications such as self-driving vehicles.
6G Performance Goals
6G will implement many different technologies to achieve its performance goals. Among them will be new topologies of overlapping cells with distributed networks of beamforming antenna controlled by AI and ML to dynamically select optimum transmission paths. Previous cellular communications were based on networks of hexagonal cells spaced far enough apart to avoid signal interference with neighboring cells. In the future, there will be overlapping cells that can be dynamically reconfigured. In addition, coverage will expand through space, sea, and high-altitude drones.
New functionality in spatial multiplexing and massive multiple-input multiple-output (mMIMO) are under investigation, including the use of reflective surfaces and metamaterials to manage signal propagation in crowded urban environments with limited LOS. 6G may employ a spatially non-orthogonal, overlapped, and dynamic topology to increase path selection. Beam control through AI and ML will help reduce intercell interference at a cost of complexity.
Besides reconfiguring networks, much of the 5G/6G focus will be on physical design of radio access front ends. Strategic design partitioning, leveraging of optimal semiconductor processes, and multi-fabric assemblies will undoubtedly be utilized, calling for a range of simulation technologies, design and manufacturing flows, and tool interoperability. These trends will require software design support for co-design and co-optimization of next-generation wireless electronic systems across multiple domains—including RF, analog, and digital simulation—aided by large-scale EM and thermal analysis, and robust design verification and signoff.
A common RFIC and system-in-package (SiP) design challenge is concerned with how RF intellectual-property (IP) blocks are subject to EM, capacitive, supply, and substrate coupling. These potentially harmful interactions result from effects that occur with high-frequency signaling. Noise, either into or out of a block, can travel through the substrate or through power bussing. Placement (floorplanning) of noise-generating and noise-sensitive blocks is design critical. Parasitic coupling capacitance within/between RF blocks, and signals routed nearby (100s to 1000s of microns away) can cause disrupted performance. Self-inductance of any nets connected to the RF block, and mutual inductance between any net within an RF block to nets in neighboring blocks, is another concern.
RF IP blocks interact much differently than digital IP blocks. Most place-and-route tools aren’t able to address the unique issues confronted in integrating RF IP. An RF engineer working with RF-aware design tools is required to successfully integrate blocks at the chip, package, or board levels. To understand the effects of parasitic coupling, either capacitive, inductive, and through the substrate, closed-form distributed transmission-line models and EM analysis are essential for RF IP development.
Using higher-frequency bands in 6G (94 GHz to 3 THz) will help reduce the size of these antennas, making efforts to shrink component footprints easier. However, the antennas, feed networks, and package interconnects will all be more susceptible to parasitics and unintended coupling (crosstalk), requiring rigorous EM analysis and design verification. This move to higher spectrum will lead to a wide range of design and integration challenges.
Need for Wireless Technology Improvements
5G deployment and related trials have shown there’s room and need for improvements to the coverage and uplink performance of NLOS environments and heavy-traffic use cases. To fully achieve the promise of this next wave of communications, we’ll need continued enhancements to guarantee the high reliability and low latency necessary to close the gap between the cyber and physical worlds. 5G evolution will concentrate on better uplinking and toward more delivery guarantees, as opposed to “best effort,” with a focus on URLLC (Fig. 4).
4. Integration of beamsteering RF front-end module technologies will improve uplink data rates for 5G/6G devices.
Achieving URLLC performance with < 1-ms latency and up to “nine nines” (99.9999999%) mission-critical reliability (for factory operational technology) for networks based on edge cloud computing will require a massive number of small-cell radio access points and distributed antennas with AI/ML controlled beamsteering. Implicit in the requirements for this vast deployment of non-orthogonal networks will be the need to drive down the cost of the beamforming antenna arrays and complex receivers, especially if faster-than-Nyquist signaling techniques are applied, while greatly improving (beyond the current best effort) uplink technology.
Conclusion
A key goal of 5G has been to expand the reach and value of wireless technology beyond the individual mobile subscriber in support of mMTC and URLLC. Expanding connectivity to include network-to-smart device communications, combined with AI and IoT, will usher in a new industrial wave and offer greater business value for both industry and society.
David Vye is Technical Marketing Director at Cadence.
September 25, 2020 at 06:17AM
https://ift.tt/2RWSWWm
Marvell to Bring 5-nm Chips to Cloud Data Centers and 5G Networks - Electronic Design
https://ift.tt/2RGyUAH
Chips
No comments:
Post a Comment