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

Current MPR Articles

November 25, 2019

• Alibaba Takes AI Performance Lead
  The Chinese Internet giant has developed its first ASIC to accelerate neural networks, posting ResNet-50 throughput that’s 10x better than Nvidia’s fastest GPUs.
  

• MLPerf Debuts Inference Benchmarks
  The MLPerf group has released a modest set of inference benchmarks and use cases. The first results include submissions from cloud-service providers (CSPs), server vendors, chip vendors, and professional-services companies.
  

November 18, 2019

• Chiplet Technology Proliferates
  Chiplets are getting lots of attention in IC-design circles, as real products are shipping in high volumes from leading vendors. Package and interface technologies are advancing chiplet-based designs.
  

• Renesas RA MCUs Strengthen Security
  Renesas Advanced (RA) microcontrollers will include the company’s first models that implement Arm’s TrustZone technology. The 32-bit RA family comprises chips with Cortex-M23 and Cortex-M4 CPUs; future products will add Cortex-M33.
  

• Editorial: Third-Party Chiplet Challenges
  The ODSA hopes to shift the chiplet market from in-house to third-party designs, but to enable this transition, it must overcome several design and manufacturing challenges.
  

• Arm Offers Bespoke Cortex-M CPUs
  Arm is opening the Arm v8-M architecture to allow custom instructions. Cortex-M33 is the first CPU to support this new feature, but all future Arm v8-M cores will as well.
  

November 11, 2019

• Emerging Memories: Why Now?
  How will an emerging memory technology rise to prominence? Both technology and market dynamics will lead to the replacement of today’s technologies with something altogether different.
  

• ARC VPX5 Sports Four-Way VLIW
  The new VPX5 is the first Synopsys ARC DSP to employ a VLIW architecture, combining a scalar CPU, a 512-bit vector/SIMD unit, and an optional vector floating-point engine. 
  

• SiFive Secures RISC-V
  SiFive recently disclosed its Shield secure-platform architecture, which includes hardware and software as well as provisioning services. Shield’s baseline capabilities are similar to those of Arm’s TrustZone.
  

November 4, 2019

• Intel’s Tremont: A Bigger Little Core
  Tremont is Intel’s newest low-power x86 CPU, boosting IPC by 30%. The new core design will debut in the Lakefield processor—a heterogeneous multicore SoC for thin notebooks and tablets.
  

• NovuMind Launches Tensor Processor
  Startup NovuMind is preparing to tape out a neural-network accelerator that it estimates will deliver 150 trillion operations per second (TOPS) while consuming just 5W, achieving impressive efficiency.
  

• BittWare Puts Achronix On Board
  A new partnership with BittWare gives Achronix another route to market for its Speedster7t FPGAs. The companies recently announced the VectorPath PCI Express accelerator card based on Speedster7t.
  

• Mythic Reveals Data-Flow Architecture
  AI startup Mythic revealed more details of its first chip, which performs matrix multiplication in the analog domain. Conventional digital circuits, however, manage data flow and other tasks.
  

• Cornami Right-Sizes Systolic Arrays
  After testing FPGA prototypes of its dynamically configurable AI processor, startup Cornami projects its first chip will run ResNet-50 at an incredible 105,000fps while burning only 30W.
  

October 28, 2019

• SiFive U8 Takes RISC-V Out of Order
  SiFive has revealed its U8-series, which builds on the 64-bit superscalar U7-series by adding out-of-order (OOO) execution, a first for commercial RISC-V intellectual property. 
  

• GrAI Matter Exploits Sparse Video
  Grai One is the first chip to implement the NeuronFlow neuromorphic architecture, which can handle spiking neural networks but excels on convolutional models for streaming video.
  

• BrainChip Akida Is a Fast Learner
  BrainChip has announced a new spiking-neural-network processor called Akida, which it plans to tape out by the end of the year. It also offers Akida’s core as licensable IP.
  

October 21, 2019

• i.MX RT1170: Fastest Flashless MCU
  NXP’s new i.MX RT1170 combines microcontroller features with high performance. Its 1.0GHz Cortex-M7 is the fastest CPU in its class, and its 400MHz Cortex-M4 coprocessor is also swift for its breed.
  

• Intel Slashes Pricing on X-Series
  Intel recently updated its Core X-series for high-end desktops and its related Xeon W for workstations. The big news, however, is dramatically reduced prices, a response to increased competitive pressure from AMD.
  

• Innovium Moves Down Market
  Teralynx 5 line ranges from 1.2Tbps to 6.4Tbps, extending the company’s portfolio to lower-cost and lower-power systems. It follows the 12.8Tbps Teralynx 7, which reached production in 1Q19.
  

October 14, 2019

• Everspin Raises MRAM Density
  The leader in discrete MRAM is shipping a 1Gb device, narrowing the density gap to DRAM. GlobalFoundries offers the Everspin technology for embedded MRAM.
  

• Editorial: The Many Faces of 5G
  5G is off to a flying start, but depending on the band support and price, different 5G devices offer higher or lower data rates and may be missing other important benefits of the new protocol.
  

• Arm Joins Bfloat Parade
  To accelerate neural-network training, Arm will support the Bfloat16 data format, which doubles throughput for the matrix-multiplication operations at the heart of most AI models.
  

October 7, 2019

• Huawei Delivers Powerful AI Training
  The company’s new Ascend 910 takes its DaVinci architecture into the top performance tier in AI training. Huawei will deliver accelerator cards, servers, and clusters incorporating the chip.
  

• Exynos 980 Integrates 5G Modem
  The Exynos 980 5G is Samsung’s first smartphone processor to integrate a 5G modem. The first processor to use Arm’s Cortex-A77, it also features Samsung’s fastest deep-learning accelerator.
  

• Ceva and Synopsys Spin More TOPS
  To support neural-network inference in autonomous vehicles and other high-performance edge devices, new multicore deep-learning accelerators can generate 100 TOPS or more.
  

September 30, 2019

• Kirin 990 First to Integrate 5G
  Huawei is the first company to fully announce a processor with an integrated 5G modem. The new flagship processor also leads in smartphone AI performance.
  

• Prepare for Copackaged Optics
  Electrical-I/O pin bandwidths are almost out of runway, and chips will soon hit the serdes wall. Copackaged optics provide a path forward by reducing reach, but they present a new set of challenges.
  

• Editorial: Intel 10nm Disappoints
  Intel recently began 10nm production, but the new technology falls short of its speed and power goals. These shortcomings affect the company’s current and future 10nm PC and server products.
  

September 23, 2019

• Tsinghua Pedals Hybrid AI Processor
  Tianjic’s novel hybrid architecture can concurrently run convolutional neural networks, multilayer perceptrons, recurrent neural networks, and other deep neural networks, as well as biologically inspired continuous-attractor and spiking neural networks.
  

• Power9 AIO Boosts DRAM Bandwidth
  IBM’s new Power9 AIO (Advanced I/O) server processor introduces the Open Memory Interface (OMI)—a superfast alternative to DDR4. Special DIMMs boost the peak bandwidth to 410GB/s, making Power9 ideal for memory-intensive applications.
  

• Qualcomm Updates Wi-Fi 6 Portfolio
  Qualcomm updated its Wi-Fi 6 products for both infrastructure and smartphones, also rebranding those respective lines as Networking Pro Series and FastConnect. Both achieved Wi-Fi Certified 6 status in the Wi-Fi Alliance’s new certification program.
  

• Broadcom Ships PCIe Gen4 Switches
  The PCI-SIG has completed its first compliance tests for PCI Express 4.0. The integrator’s list includes Broadcom’s PEX88000 PCIe Gen4 switch chip, which is now in production.
  

September 16, 2019

• HPE Reveals Gen-Z Fabric Chipset
  At Hot Chips, HPE disclosed industry-first Gen-Z bridge and switch chips with copackaged optics. They target exascale systems rather than mainstream servers.
  

• Unisoc Hits Mid-Premium Segment
  Raising its sights, Unisoc disclosed two new smartphone processors. The Tiger T710 offers industry-leading AI performance, while the Tiger T618 focuses on improved image processing.
  

September 9, 2019

• Mellanox Right-Sizes Smart-NIC SoC
  Although it expects to become part of Nvidia by year-end, Mellanox is pressing forward with its server-connectivity roadmap, announcing second-generation BlueField processors as well as a new Ethernet controller optimized for cloud data centers.
  

• Comet Lake Walks Over Ice Lake
  The 14nm Comet Lake is a 10th Generation alternative to the 10nm Ice Lake processor for ultrathin and hybrid laptops, offering up to six CPU cores and faster Wi-Fi than last year’s models.
  

• Cypress Tracks Indoor Locations
  Cypress’s new CYW5459 is a multimode Bluetooth+Wi-Fi combo chip that’s especially well suited to indoor asset tracking for consumer electronics, IoT, smart-home devices, and automotive.
  

September 2, 2019

• Cerebras Breaks the Reticle Barrier
  Designed for training neural networks, the startup’s first processor is by far the largest integrated circuit ever manufactured, measuring 215mm on a side and encompassing 1.2 trillion transistors.
  

• Spring Hill Sets AI Efficiency Mark
  Intel’s Nervana NNP-I accelerator is rated at 4.8 TOPS per watt and is 5x more power efficient than Nvidia’s best products for ResNet-50 inference. It fits into a tiny OCP-compliant M.2 card.
  

• Ayar Labs Realizes In-Package Optics
  The startup hopes to commercialize silicon-photonic chiplets next year, enabling customer MCMs with in-package optics. It’s working with GlobalFoundries for process technology and with Intel for MCM technology and product prototypes.
  

More Articles

Purchase an Article