Linley Newsletter: April 5, 2018

 weSRCH's Best of the Internet Award

Linley Newsletter

(Formerly Processor Watch, Linley Wire, and Linley on Mobile)

Please feel free to forward this to your colleagues

Issue #594

April 5, 2018

Independent Analysis of Microprocessors and the Semiconductor Industry

Editor: Tom R. Halfhill

Contributors: Linley Gwennap, Mike Demler, Bob Wheeler

In This Issue:

- Mali-G52 Boosts Mainstream Phones

- Abee Adds AI to Wearables Processor

- Microsemi Pushes OTN to Terabits

Mali-G52 Boosts Mainstream Phones

By Mike Demler

Arm's new Mali graphics and multimedia cores target the mainstream smartphone market. The Mali-G52 improves area, performance, and power-efficiency compared with the Mali-G51, the first mainstream Bifrost GPU. In addition to aligning GPU/VPU product numbering, the Mali-V52 video codec supports decoding/encoding at 4K resolution. Completing the suite is the Mali-D51, a new display processor core that offloads composition, rotation, scaling, and other tasks from the GPU.

Dedicated neural-network inference engines are now standard for running machine-learning applications on premium smartphone processors, but the Mali-G52 offers a lower-cost alternative by running these applications on a GPU. Relative to its predecessor, the new design doubles the execution-engine width from four to eight lanes, each supporting four single-cycle multiply-accumulators (MACs). According to the company's estimates, the lane expansion increases the GPU die area by just 22%.

To further boost neural-network performance, the G52 moves INT8 dot-product calculations from the floating-point pipeline to dedicated fixed-point units. For MAC-intensive convolution layers, Arm estimates these improvements raise inference-engine performance by 3.6x compared with the Mali-G51.

The company is also offering a smaller Bifrost GPU for the low-cost market. The new Mali-G31 gives designers a choice of single- or dual-pixel shaders, along with an option to use one or two execution units per shader. In a 16nm process, we estimate the tiny single-execution-unit G31 consumes roughly 1.0mm^2. It's the smallest Mali GPU to support the OpenGL ES 3.2 and Khronos Vulkan APIs. The cores are available now for licensing, and we expect the first smartphones to launch in 1H19.

Microprocessor Report subscribers can access the full article:

Abee Adds AI to Wearables Processor

By Linley Gwennap

Fresh off a successful IPO, the company that shipped the most wearable devices last year is preparing to sample its first custom processor. No, it isn't Apple or Fitbit but rather Huami (pronounced "wha-me"), maker of the popular Mi Band. Until recently, few people outside of China have heard of Huami, which is a supplier to Xiaomi, a top Chinese phone vendor that sells the Mi Band. Huami in turn has created a subsidiary, Abee Semiconductor, to develop custom processors for its wearable devices.

The subsidiary's first chip, simply called the Abee1, targets fitness bands and midrange smartwatches. It features a scalar RISC-V CPU along with a simple AI accelerator that the company says produces a 23x power savings when running common neural networks. The 55nm SoC also includes basic microcontroller peripherals and on-chip SRAM, although it requires external flash memory. The tiny chip consumes just 41mW at 280MHz and less than 0.03mW when monitoring sensors.

Abee Semiconductor was founded in June 2016 by CEO/CTO Nhon Quach, formerly an SoC architect at Intel, Microsoft (Xbox), and Qualcomm. It's currently funded solely by Huami, which provided an initial $6 million stake with a $20 million total commitment, although it hopes to attract outside investors. Abee has only seven employees, with its headquarters in Silicon Valley and the engineering staff in Taiwan. To develop an SoC using such a small team, it contracted with Faraday for ASIC-design services and recently taped out the chip in UMC's 55nm process. The startup plans to sell the chip to other customers in addition to Huami, subject to the parent's approval.

Microprocessor Report subscribers can access the full article:

Microsemi Pushes OTN to Terabits

By Bob Wheeler

Microsemi's Digi-G5 is the industry's first OTN framer/mapper chip to handle 400G Ethernet as well as Flex Ethernet (FlexE). Due to sample this quarter, it combines up to 1.2Tbps of bandwidth and unprecedented flexibility. It follows by three years the 400Gbps Digi-G4, which handled OTN rates as high as 100Gbps. These products come from the company's 2016 acquisition of PMC-Sierra.

The Digi-G5 starts with obvious improvements such as a move from 28nm to 16nm technology and support for 400GbE-over-OTN. It also adopts 56Gbps PAM4 long-reach serdes, which serve both client- and line-side interfaces. For standard Ethernet rates, it handles 2x400GbE, 12x100GbE, and 24x25GbE. The new chip also, however, adds flexible port rates enabled by new standards. Specifically, it implements FlexE on the client side, aggregating as many as six 100GbE ports into a single flow of up to 600Gbps. On the line side, the chip handles the ITU's new OTUCn scheme and flexible OTN short-reach interface (FlexO), which likewise aggregate Nx100Gbps channels.

Like its predecessor, the Digi-G5 offers subwavelength AES-256 encryption, allowing each customer's traffic to be encrypted end to end as it's transported over shared wavelengths. It supports up to 600Gbps of wire-speed encryption, which increases latency by only 180ns. The new chip adds an ARM CPU (of undisclosed type) to offload statistics collection, protection switching, and other real-time tasks from the external host CPU. Its SDK implements a hardware abstraction layer that exposes a high-level API to the OEM software stack.

Historically, five major vendors developed optical-transport framer/mapper chips, but now only Microsemi remains active. At the same time, Microchip is acquiring Microsemi, with the deal expected to close this quarter. Lacking merchant competition, the Digi-G5 positions the company to continue its share gains as OEMs roll out next-generation OTN systems.

Microprocessor Report subscribers can access the full article:

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:

Domain: Electronics
Category: Semiconductors

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