Linley Newsletter: January 26, 2017

 weSRCH's Best of the Internet Award
  26th-Jan-2017
 548

Linley Newsletter
(Formerly Processor Watch, Linley Wire, and Linley on Mobile)
Issue #532
January 26, 2017

Independent Analysis of Microprocessors and the Semiconductor Industry

Editor: Tom R. Halfhill
Contributors: Loyd Case, Linley Gwennap, David Kanter, Bob Wheeler

In This Issue:

- Year in Review: Ethernet Accelerates As Wireless Slows
- Analysts' Choice Winners for 2016
- Ambiq Rides Quadratic Power Curve
- AMD Hones Learning Instincts

Coming Soon: "A Guide to CPU Cores and Processor IP"

We are pleased to announce the upcoming publication of the Fifth Edition of our report "A Guide to CPU Cores and Processor IP." This report analyzes IP vendors and their core products, probing their strengths and weaknesses and presenting key details in a consistent, easy-to-compare fashion. We focus on CPU, GPU (graphics), and network-on-a-chip (NoC) IP, examining performance, die area, power, and other relevant parameters. Because selecting a processor core has long-term software implications, we also discuss the vendors' product roadmaps.

The report provides in-depth coverage of licensable CPU and GPU cores including ARM's Cortex and Mali cores, Imagination Technology's PowerVR GPUs and MIPS CPUs, Synopsys ARC CPUs, and Cadence Xtensa CPUs. Also covered are IP cores from Adapteva, Andes Technology, Beyond Semiconductor, Ceva, Cortus, SiFive/RISC-V, and VeriSilicon/Vivante, plus NoC IP from Arteris, NetSpeed, and Sonics. These cores can be used in chips for automotive, consumer devices, smartphones, networking equipment, servers, microcontrollers, and PCs.

A detailed description, including the preliminary table of contents, can be found on our web site:
http://www.linleygroup.com/cpu-ip

Prepublication offer: Order by February 9, 2017 and save $300 off the list price.

Year in Review: Ethernet Accelerates As Wireless Slows
By Bob Wheeler

As many companies digested 2015 acquisitions, business mergers slowed in 2016. Instead, vendors refocused on delivering new products, particularly for data-center networks. Another market garnering attention was IoT, with low-power wireless modems seeing the most activity. Next-generation cellular technology, 5G, climbed the hype curve.

Incumbents Broadcom and Marvell each introduced multiple new Ethernet switch chips in 2016. The former primarily addressed public clouds and carrier networks, whereas the latter focused on private clouds and enterprise networks. Barefoot Networks joined the ranks of 100GbE switch-chip vendors, and Cavium brought its first such chip to production. Using FinFET technology, Barefoot and Broadcom unveiled the industry's first 6.4Tbps switch chips.

Ethernet PHY vendors laid the groundwork for PAM4 deployments, first at 50Gbps per lane but with 100Gbps per lane close behind. Interestingly, chipmakers addressed applications ranging from 3-meter server-access links to 80-kilometer data-center interconnects (DCIs). Server-adapter (NIC) vendors introduced a wide variety of products for 25GbE and 100GbE. Consolidation hit the server-connectivity market as Cavium acquired QLogic, which sells both Fibre Channel HBAs and Ethernet NICs.

LTE build-outs are complete or slowing in many regions, so wireless operators and suppliers delivered new IoT products and services while developing 5G prototypes in 2016. The 3GPP completed work on LTE categories optimized for battery-powered IoT devices, providing a low-bandwidth replacement for aging 2G technology. Operators pushed for rapid progress on 5G technology; vendors announced development platforms and preannounced modem chips.

Whereas China's LTE deployment is winding down, its data centers are growing rapidly. At the same time, 2017 will undoubtedly see changes in U.S.-China trade relations. These changes could affect U.S. vendors hoping to capitalize on the growing Chinese market as well as Chinese companies looking to expand through acquisitions.

Microprocessor Report subscribers can access the full article:
http://www.linleygroup.com/mpr/article.php?id=11745

Analysts' Choice Winners for 2016
By Linley Gwennap

To recognize the top semiconductor offerings of the year, The Linley Group presents its 2016 Analysts' Choice Awards. This year's winner for Best Server Processor is Intel's Xeon Phi 7200 family (formerly Knights Landing), which offers an innovative and novel system architecture for high-performance computing (HPC). Because the CPUs are fully x86 compatible, the Xeon Phi 7200 is bootable and can run an operating system. The cumulative benefit of these changes is that the processor comprises a nearly complete HPC node, so it avoids the system-level complexity of a coprocessor-compute model.

In the embedded-processor category, Marvell's Armada 88SP1000 edged out NXP's LS2088A. Setting Marvell apart are its integration, performance, and low power. For the 88SPxxx-series, Marvell combined SiP (system-in-package) with its MoChi (modular chip) technology to offer three lines.

We selected Cavium's Xpliant Ethernet switch as our best networking chip. The Xpliant CNX88091 stands out thanks to its customer programmability. Although it trailed Broadcom's Tomahawk in reaching production, multiple OEMs have announced systems with unique features enabled by the Xpliant architecture.

We chose the Qualcomm Snapdragon 820 as our Best Mobile Processor for best addressing the needs of premium smartphones. Its custom-designed Kryo CPU delivers better single-core performance than any merchant smartphone processor, and its custom Adreno GPU leads in graphics performance. The 820's LTE modem is rated at 600Mbps, another market-leading speed.

For the best mobile chip (not including processors), we chose the SmarTuner from Cavendish Technologies. The SmarTuner enables capacitive aperture tuning, increasing antenna efficiency, enhancing reception, and saving power. Its 32 levels handle a wide range of LTE bands and antenna types.

Our choice of Best Processor IP is the Synopsys ARC SEM CPU family. The SEM110 and SEM120D address the critical need for more-secure IoT clients. They employ a tamper-resistant pipeline that uses address scrambling along with in-line encryption of data and instructions.

Finally, the RISC-V ("risk-five") instruction-set architecture (ISA) wins our Best Technology Award. RISC-V is a general-purpose architecture that's BSD licensed, extensible, and royalty free. It's flexible enough for all markets, from microcontrollers to server processors. The ISA has attracted support from industry leaders such as Google and Oracle as well as academic partners.

Microprocessor Report subscribers can access the full article:
http://www.linleygroup.com/mpr/article.php?id=11746

Ambiq Rides Quadratic Power Curve
By Loyd Case

Ambiq Micro's new Apollo 2 processor uses a 48MHz Cortex-M4F CPU that consumes less than 10 microamps per megahertz when executing from flash memory or embedded SRAM. Mainstream MCUs that sport the same ARM CPU require 80 microamps per megahertz or more. Ambiq reaches its bold power goals by running much of the SoC at subthreshold voltages. Subthreshold circuits drive transistors at voltages less than what's necessary to create a conducting path between the source and drain. Although it employs standard supply voltages, Ambiq has tuned portions of both the logic and analog circuitry to run at about 0.5V internally. This voltage reduction extends battery life for IoT and wearable devices. The new MCU first sampled in 2Q16 and is scheduled to begin production in 1H17.

Ambiq developed a proprietary ultra-low-voltage approach called Subthreshold Power Optimized Technology (SPOT), which it used in several products before Apollo 2. Among those products are a pair of real-time clocks (one of which incorporates a PMIC) and the original Apollo MCU, which employs a 24MHz Cortex-M4F CPU.

The company implemented the logic (including the CPU's physical layout), analog components, and memory using subthreshold circuits. Running at these extremely low voltages greatly reduces power consumption, which decreases quadratically with voltage. Similar microcontrollers from NXP and STMicroelectronics require 10-30x more operational current.

Microprocessor Report subscribers can access the full article:
http://www.linleygroup.com/mpr/article.php?id=11747

AMD Hones Learning Instincts
By David Kanter

Leaping into the hot machine-learning market, AMD unveiled the Radeon Instinct GPU line, topped by the MI25 training accelerator and the MI6 and MI8 inference accelerators. AMD has not disclosed their production availability, but the MI6 and MI8 are based on the existing Polaris and Fiji microarchitectures, respectively, and should arrive in 2Q17. The MI25 is based on the forthcoming Vega microarchitecture and so could take a bit longer; we expect it to arrive in 3Q17.

All Radeon Instinct accelerators eschew fans and use passive cooling appropriate for the data center. The product-naming system is remarkably sensible and appears to reflect each model's computational throughput for half-precision floating-point (FP) operations. The designs feature a variant of Single Root I/O Virtualization (SR-IOV) with a 64-bit PCI Express address space and multi-GPU support using that space.

The MI6 is based on the 14nm Polaris 10 architecture, which appears in the Radeon RX 480. It features 36 shader cores and a 256-bit GDDR5 interface to 16GB of memory. When operating at 1.1GHz, it draws 150W and delivers 5.7Tflop/s of single- or half-precision compute throughput. The more powerful MI8 is based on the Nano version of the older 28nm Fiji architecture, packing 64 cores and 4GB of High Bandwidth Memory (HBM). Despite the older process, the more efficient memory interface enables 8.2Tflop/s of single- or half-precision performance -- 44% higher than the MI6 at the cost of just 17% more power. But both accelerators deliver only about half the power efficiency of the MI25.

Microprocessor Report subscribers can access the full article:
http://www.linleygroup.com/mpr/article.php?id=11750

About Linley Newsletter

Linley Newsletter is a free electronic newsletter that reports and analyzes advances in microprocessors, networking chips, and mobile-communications chips. It is published by The Linley Group and consolidates our previous electronic newsletters: Processor Watch, Linley Wire, and Linley on Mobile. To subscribe, please visit:
http://www.linleygroup.com/newsletters/newsletter_subscribe.php

Our previous subscription newsletters -- Microprocessor Report, Networking Report, and Mobile Chip Report -- are now consolidated in a single subscription newsletter: Microprocessor Report. MPR publishes articles with significantly more detail on the subjects covered in our free electronic newsletter. For subscription information, please contact us at 408.270.3772; cs@linleygroup.com.

Back issues: http://www.linleygroup.com/mpr/archive.php?j=TLG

View our privacy policy here: http://www.linleygroup.com/privacy.php

Domain: Electronics
Category: Semiconductors
SEMICONDUCTOR ANALYTICS

Recent Newsletters

Linley Newsletter: August 8, 2019

Linley Newsletter Please feel free to forward this to your colleagues Issue #664 August 8, 2019 Independent Analysis of Microprocessors and the Semiconductor Industry E

08 August, 2019

Linley Newsletter: August 1, 2019

Linley Newsletter Please feel free to forward this to your colleagues Issue #663 August 1, 2019 Independent Analysis of Microprocessors and the Semiconductor Industry E

01 August, 2019

Linley Newsletter: July 25, 2019

Linley Newsletter Please feel free to forward this to your colleagues Issue #662 July 25, 2019 Independent Analysis of Microprocessors and the Semiconductor Industry

25 July, 2019