» Archived Articles | Login | Activate My Subscription | Subscribe | Free Sample Articles | Purchase an Article

Current MPR Articles

August 26, 2019

• Intel Spring Crest Achieves Nervana
  The Nervana NNP-T is the company’s first dedicated accelerator for neural-network training, offering 119 TOPS using the emerging Bfloat16 format. Intel plans to announce products later this year.
  

• Alibaba’s Ouroboros Is a Fast Talker
  To provide improve voice interfaces in embedded devices, Alibaba is developing an AI-enabled speech-synthesizer chip that can run DeepMind’s WaveNet model using as little as 200mW. 
  

• Upmem Embeds Processors in DRAM
  French startup Upmem is tackling the processor-memory bottleneck by integrating processor cores into DRAM chips. The custom memory modules fit into standard DIMM slots.
  

August 19, 2019

• MediaTek Helio G90 Ups Its Game
  To target gamers, the new Helio G90 and the performance-binned Helio G90T improve the CPU, display, graphics, and wireless-connectivity performance compared with the Helio P90.
  

• Editorial: 5G to the Rescue?
  Smartphone shipments are trending down, particularly in the premium segment. Even the 5G transition won’t be enough to prop up prices over the next few years.
  

• Snapdragon 855+ Aims at Gaming
  In a midyear upgrade, Qualcomm rolled out a new version of its flagship processor that provides greater CPU performance and faster graphics. It debuts in the Asus ROG Phone 2 for mobile gamers.
  

August 12, 2019

• AMD Rome Ruins Intel Hegemony
  The second-generation Epyc processors have more cores, more memory, faster I/O, and smaller transistors than Intel’s newest Xeons. The 64-core Rome outperforms the fastest Xeons and sells for less.
  

• Ice Lake Debuts 10nm, New Cores
  Intel has debuted its long-delayed 10nm technology in the new U- and Y-series notebook processors. Clock speeds suffer, but the new Sunny Cove CPU and expanded Gen11 GPU boost performance.
  

• Intel's Loihi Mimics Biological Neurons
  Intel’s Loihi is a spiking-neural-network processor comprising 128 programmable cores arranged in an 8x16 mesh-connected array. Each core can model up to 1,024 biological neurons and 128 million synapse connections.
  

August 5, 2019

• An Inside Look at AI Accelerators
  Our die analysis exposes the core counts, memory size, and area that leading mobile processors devote to AI acceleration. We break down the current AI-Benchmark results to isolate accelerator performance.
  

• Arm IP Licensing Gets More Flexible
  Arm’s new Flexible Access program enables customers to defer licensing fees until tapeout. The program includes selected Cortex-A, selected Cortex-R, and all Cortex-M CPUs, as well as Mali GPUs.
  

July 29, 2019

• Huawei Takes Lead in Smartphone AI
  The Kirin 810 processor targets the highly competitive mid-premium smartphone tier, but the chip features 7nm technology and a new in-house accelerator that offers premium-level AI performance.
  

• ST Accelerates Motor Control
  The STM32G4-series from STMicroelectronics features unique math accelerators for controlling electric motors. These Arm-based MCUs are also well equipped with mixed-signal peripherals.
  

• Helio P65 Upgrades Midrange Phones
  MediaTek’s new midrange application processor adopts the popular 2+6 CPU configuration using Arm’s Cortex-A75 and Cortex-A55. It also has an Arm Mali-G52 GPU, which boosts gaming and AI performance.
  

July 22, 2019

• CXL Enters Coherent-Accelerator War
  The Intel-developed interconnect targets data-center accelerators including GPUs, FPGAs, and ASICs. CXL competes directly with CCIX and OpenCAPI, which came earlier but garnered few chip designs.
  

• Qualcomm Upgrades Low-End Phones
  Qualcomm’s new 215 mobile platform targets smartphones costing $75 to $125 in emerging markets. The entry-level chipset now has 64-bit CPUs, 802.11ac, NFC, dual ISPs, and support for two SIMs.
  

July 15, 2019

• Rockchip Accelerates AI for IoT
  Rockchip’s new RK1808 is a neural-network processor designed for IoT and similar low-power devices. The RK3399Pro is an upgraded version of the RK3399, which the company previously supplied for Google Chromebooks.
  

• ST’s Speed-Demon MCUs
  STMicroelectronics is shipping exceptionally fast microcontrollers that can reach 480MHz. The new dual-core STM32H7 MCUs add a Cortex-M4F to assist the Cortex-M7 main CPU. 
  

• Editorial: Deal Fever Hits Comms Semis
  Networking chips have seen a recent acquisition flurry. Unlike previous scale-driven deals, most of this activity has been more strategic—it reflects portfolio expansion rather than simple market consolidation.
  

July 8, 2019

• AMD Navi Debuts RDNA Architecture
  RDNA is AMD’s first major new graphics architecture in seven years. Its first incarnation is the gaming-focused Radeon RX 5700-series, implementing a new programming model and a comprehensive set of upgrades.
  

• Mali-G76 GPU Boosts Area Efficiency
  Arm touted a 30% improvement in performance per unit area for its Mali-G76 design, but our analysis of the initial G76-based products reveals nearly a 40% gain.
  

• Microchip Brings FlexE to Routers
  Microchip has elevated the former-Microsemi Meta line from the shadow of Digi OTN products to purpose-built Ethernet PHYs. The 1.2Tbps Meta-DX1 family handles 400G Ethernet combined with advanced features.
  

July 1, 2019

• Zen 2 Boosts Ryzen Performance
  AMD’s new PC processors are based on the 7nm Matisse design, which includes the new Zen 2 microarchitecture. They outperform comparable Intel products on multithreaded benchmarks.
  

• Aspinity Detects Voice in Analog Nets
  Aspinity’s Reconfigurable Analog Modular Processor (Ramp) provides always-on capability in low-power voice-activated devices, drawing just 10 microamps during active operation.
  

• Intel Refreshes Xeon E Family
  Intel’s new Xeon E-2200 processors refresh the E-2100 series, targeting desktop and mobile workstations and small-office servers. A few embedded models offer lower power and longer availability.
  

June 24, 2019

• Broadcom Samples Trident 4 Switch
  Broadcom’s first 7nm switch chip is also its first to offer compiler programmability. It marks the debut of the company’s Network Programming Language (NPL).
  

• Hailo Illuminates Low-Power AI Chip
  Israeli startup Hailo is entering the highly competitive AI inference-engine market with a chip that runs the entire ResNet-50 neural network using only on-chip SRAM and consuming just 1.5W.
  

• Knowles Adds AI to IA8201
  The IA8201 adds a midrange audio processor to the Knowles AISonic line, which includes neural-network functions for voice recognition in smart speakers, appliances, and other consumer gear.
  

• Unisoc Launches Tiger T310
  The Tiger 310 is the first low-cost smartphone processor to include a big CPU to boost performance, helping the four-core design compete against eight-core products.
  

June 17, 2019

• Habana Offers Gaudi for AI Training
  The startup expects its second AI chip to deliver better performance than Nvidia’s V100 GPU at half the operating power. It plans to sample Gaudi-based cards and systems this year.
  

• MediaTek AI Engine Earns Top Marks
  The Helio P90 integrates a custom deep-learning accelerator, enabling it to take the top spot on the new AI Benchmark. MediaTek’s forthcoming 5G SoC will feature an enhanced version of this accelerator.
  

• Silicon Labs Upgrades Wireless MCUs
  New Wireless Gecko Series 2 microcontrollers for IoT offer stronger security, enhanced radios, and Arm Cortex-A33 CPUs. These EFR32-series chips also feature the smallest packages in their class.
  

June 10, 2019

• DSP-IP Vendors Target 3D Navigation
  DSPs are well suited to simultaneous localization and mapping (SLAM). Cadence’s new Vision Q7 improves SLAM performance, and Ceva is also offering a new SLAM SDK that works with its DSPs. 
  

• Editorial: AI Needs Flexibility…For Now
  AI accelerators must handle the myriad types of neural networks available now and in the future, but over time, chip designers will focus on efficiently executing a smaller set of models.
  

June 3, 2019

• Arm Mali-G77 Goes to Valhall
  Arm’s new Mali-G77 GPU boosts area and power efficiency by 30% and machine-learning performance by 60%. The new Mali-D77 display unit includes features that target VR headsets.
  

• Achronix Debuts FPGAs for AI
  The new Speedster7t FPGA family from Achronix competes with Intel and Xilinx by offering faster DSP blocks optimized for machine learning, optional GDDR6 memory, 400G Ethernet, PCI Express Gen5, and novel internal routing.
  

• Arm CPUs Approach Intel Complexity
  Our die-photo analysis reveals that the largest Arm-compatible CPUs, such as the Samsung M4, consume up to 70% of the area of the Intel Skylake CPU after adjusting for IC-process differences.
  

May 27, 2019

• Cortex-A77 Improves IPC
  Arm’s next-generation big core offers more per-clock performance than Cortex-A76 by increasing the reorder window and by adding two new function units and a level-zero cache.
  

• Arm Machine-Learning Core Scales Up
  Arm’s machine-learning processor core is now available for general licensing, although it has yet to receive a brand name such as Cortex or Mali. A single core delivers up to four trillion operations per second. 
  

• Xilinx Enters Smart-NIC Market
  With its $400 million acquisition of Solarflare, Xilinx is bringing important server-networking intellectual property in house. The move enables it to offer FPGA-based network interface cards.
  

More Articles

Purchase an Article