IBM Optics Go CMOS

Release time:2017-06-09
source:EE Times

Researchers from IBM this week are describing a breakthrough in 60-gigabit-per-second (Gb/s) optical interconnect that the company claims will lead to broad replacement of costlier 56 Gb/s copper interconnects.

At the 2017 Symposia on VLSI Technology and Circuits, in Kyoto, Japan, scientists from IBM Research in Zurich will describe an inexpensive 60 Gb/s optical receiver that is expected to be followed next year by a matching optical transmitter. Together, the two devices will form a complete optical-transceiver built in CMOS at costs that the company expects to be lower than the costs of a copper interconnect.


IBM reports a low power implementation of a 60Gb/s optical receiver (RX) cast in 14-nanometer CMOS finFETs with high jitter tolerance using digital clock and data recovery. 

(Source: IBM)

"We are developing a single lane 60-Gigabit per second optical receiver with non-return to zero (NRZ) signaling targeting low cost multi-mode vertical-cavity surface-emitting laser (VCSEL) based links," Alessandro Cevrero, an engineer at IBM,  told EE Times in advance of the symposium. 

"The power is way lower than our competitors, ~120mW for the receiver and eventually below 300mW for the full transceiver," Cevrero said. "Also, its compact CMOS footprint and low power consumption means it can be moved closer to the processor or switch chip and eventually even be put in the same package or even on processor chip die, providing high bandwidth connectivity directly from the processor or switch chip spanning up to 100-meters. This covers links from processor-to-processor, processor-to-memory, from drawer-to-drawer inside a rack and from a rack to a tier-1 Internet switch."

Alessandro Cevrero, an electrical engineer at IBM Research in Zurich, demonstrates IBM's low-cost silicon photonics receiver. 
(Source: IBM)

Cevrero said that implementing the devices in CMOS enabled IBM to essential double transmission speed, essentially cutting the costs per Gigabit per second by two. "Some people believed that a SiGe solution was required to achieve good optical sensitivity at data rates above 32Gb/s," Cevero said. "Our work demonstrates that CMOS can achieve the same sensitivity, but at much lower power consumption.”

The 60 Gb/s optical link IBM demonstrated still depends on discrete III-V photodetectors (for the receiver) and discrete III-V lasers (for the transmitter) together forming a transceiver that is otherwise all-CMOS. Others, such as Intel (which offers a 25-Gigabit per second optical transceiver), use silicon photonics to modulate the light from a III-V lasers. Intel combines four such channels to achieve 100-Gbits per second today, but at much higher cost and power consumption, according to IBM. Intel, however, is shooting for the same goal as IBM by 2020.

IBM's current prototype runs at a wavelength of 850 nanometers, which is the standard wavelength for VCSEL-based multi-mode optical links, making it suitable for processor-to-memory, processor-to-processor and server-to-server communications. Once the complete transceiver is demonstrated later this year or early 2018, the price crossover point will have been reached, according to Thomas Toifl, manager of the high-speed Interconnects group at IBM Research in Zurich.

The entire 60-Gigabit per second optical receiver fits on a small CMOS die, or could be incorporated into the edge of a processor, and consumes only 117 milliwatts.
(Source: IBM)

"So far, optical links were always pushed out due to their higher costs, but now we have reached the point where optics are at the same price as electrical links," Toifl told EE Times in advance of the VLSI Symposium. "Electrical links, however, need complex equalization when we go to higher data rates, and hence require more power. Also, their distance is limited to about two meters of cable compared to 100 meters for our optical solution."

Toifl also claimed IBM's "breakthrough" CMOS photonics technology provides superior sensitivity ( -9dBm) and is ideal for the high throughput requirements of cloud computing. The team also claims to have pushed its existing CMOS circuitry to 70+ Gigabits per second already, but is waiting for the III-V photodiodes and vertical-cavity surface-emitting lasers to catch up before publicizing it.

IBM has already demonstrated graphene photodetectors on silicon-on-insulator substrates and silicon-germanium lasers, but Toifl claims their next-generation all-CMOS transceivers will not require germanium, making them exceptionally cool running.

Online messageinquiry

IBM Research has reported an algorithmic breakthrough for deep learning that comes close to achieving the holy grail of ideal scaling efficiency: Its new distributed deep-learning (DDL) software enables a nearly linear speedup with each added processor (see figure). The development is intended to achieve similar speedups for each server added to IBM’s DDL algorithm. The aim “is to reduce the wait time associated with deep-learning training from days or hours to minutes or seconds,” according to IBM fellow and Think Blogger Hillery Hunter, director of the Accelerated Cognitive Infrastructure group at IBM Research. Hunter notes in a blog post on the development that “most popular deep-learning frameworks scale to multiple GPUs in a server, but not to multiple servers with GPUs.” The IBM team “wrote software and algorithms that automate and optimize the parallelization of this very large and complex computing task across hundreds of GPU accelerators attached to dozens of servers,” Hunter adds. Graph shows linear speedup to deep-learning algorithms as GPUs are added. (Source: IBM) IBM claims test results of 95 percent scaling efficiency for up to 256 Nvidia Tesla P100 GPUs added to a single server using the open-source Caffe deep-learning framework. The results were calculated for image recognition learning but are expected to apply to similar learning tasks. IBM achieved the nearly linear scaling efficiency in 50 minutes of training time. Facebook Inc. previously achieved 89 percent efficiency in 60 minutes of training time on the same data set. IBM is also claiming a validation accuracy record of 33.8 percent on 7.5 million images in just seven hours of training on the ImageNet-22k data set, compared with Microsoft Corp.’s previous record of 29.8 percent accuracy in 10 days of training on the same data set. IBM’s processor was its PowerAI platform — a 64-node Power8 cluster (plus the 256 Nvidia GPUs) — providing more than 2 petaflops of single-precision floating-point performance. The company is making its DDL suite available free to any PowerAI platform user. It is also offering third-party developers a variety of application programming interfaces to let them select the underlying algorithms that are most relevant to their application.
2017-08-14 00:00 reading:344
IBM researchers have set a tape areal-density record of 201 gigabytes per square inch — 20 times the areal density of current commercial tape drives — enabling a single palm-sized cartridge to hold 330 terabytes of uncompressed data. IBM Research and Sony Storage Media Solutions, which developed the nano-grained sputtered tape used for the demonstration prototype, described the achievement in Tsukuba, Japan, today (Aug. 2) at The Magnetic Recording Conference (TMRC 2017). IBM scientist Mark Lantz holds a 1-square-inch piece of sputtered tape that can hold 201 Gbytes.  (Source: IBM)   IBM Research scientists have achieved a new record in tape storage on a prototype sputtered magnetic tape developed by Sony Storage Media Solutions.  (Source: IBM) Tape was invented more than 60 years ago and has repeatedly been deemed obsolete, but it remains the dominant method for storing cold data — data that is infrequently accessed but must be maintained, such as tax documents and health care records. The Big Data era has seen a resurgence in popularity for tape, which is valued for its small size and low cost relative to other storage alternatives as well as for its ability to store not just backup and archival data, but also the massive sensor and transactional data streams going up to the cloud. Indeed, business at IBM's tape storage unit grew by 8 percent last year, according to Gartner Inc. IBM has set five tape storage records since 2006.  (Source: IBM) An increase in areal recording density simply means that less space is needed to store massive amounts of seldomly accessed information. IBM has broken the areal density record for tape storage five times since 2006 (see table), with most of the breakthroughs ending up in commercial products. Although the IBM-Sony's demonstration at TMRC 2017 is a prototype, IBM suggested that a commercial product could arrive next year. The prototype pairs Sony’s nano-grained magnetic tape technology with IBM Research Zurich’s newly developed write/read heads and IBM’s next-generation servo and signal processing technologies.  (Source: IBM) Sony achieved nano-granularity by sputtering vertically oriented, 7-nanometer magnetic grains on a tape substrate topped with a protective layer and a permanent lubricant. IBM created signal-processing algorithms that use noise-predictive detection to enable a linear density of 818,000 bits per inch of tape using an ultra-narrow, 48-nm tunneling magnetoresistive head. The ultra-narrow tracks enable a thirteenfold increase in track density over IBM’s previous generation, to 246,200 tracks per inch with a bit-error rate of <1e-20, the researchers reported.
2017-08-04 00:00 reading:431
IBM researchers and their partners have developed a new transistor architecture based on stacked silicon nanosheets that they believe will make FinFETs obsolete at the 5nm node.  The architecture, which was described Monday (June 5) at the 2017 Symposia on VLSI Technology and Circuits conference in Kyoto, Japan, is the culmination of 10 years of research on nanosheets by IBM, its Research Alliance partners GlobalFoundries and Samsung, and equipment suppliers. Compared to FinFETs, the new architecture consumes far less power, according to the researchers. The IBM/GlobalFoundries/Samsung revolutionary 5-nanometer nano sheet transistors, with quadruple all-around-gates should give smartphones 2-to-3 day battery life.  (Source: IBM) The Alliance breakthrough should enable battery powered devices like smartphones and other mobile devices to run for 2-to-3 days on a single charge, as well as boost performance of artificial intelligence (AI), virtual reality and even supercomputers, they say. Less than two years after developing 7nm test chips with 20 billion transistors, the researchers say they have paved the way for 30 billion transistors on a fingernail-sized chip with quadruple all-around nanowire gates. Test results indicate a 40 percent boost in performance (at the same power as 7nm FinFETs) or up to a 75 percent savings in power compared with today's advanced 10nm transistors. IBM Research scientist Nicolas Loubet holds a wafer of chips using 5nm silicon nanosheet transistors manufactured using an industry-first process. (Photo Credit: Connie Zhou)  (Source: IBM) According to IBM, the new 5nm breakthrough to more performance will boost its cognitive computing efforts as well as everybody's efforts toward higher-throughput cloud computing and deep learning, along with lower power and longer battery life for all mobile Internet-of-Things (IoT) devices. IBM scientists at the SUNY Polytechnic Institute Colleges of Nanoscale Science and Engineering’s NanoTech Complex in Albany, NY prepare test wafers with 5nm silicon nanosheet transistors to test an industry-first process of building silicon nanosheets. (Photo credit: Connie Zhou)  (Source: SUNY) To achieve the breakthrough the Research Alliance had to overcome the problems plaguing EUV (extreme ultraviolet) lithography, which was already on its roadmap for producing 7nm FinFETs. Beside the shorter wavelength advantage of EUV, the Research Alliance also found ways to continuously adjust the width of its nanosheets in both the chip design and manufacturing process phases. This fine-tuning of performance versus power tradeoffs is impossible for FinFETs, which are constrained by their fin height, rendering them unable to increase current flow for higher performance when scaled to 5nm, according to the researchers. Science and Engineering’s NanoTech Complex in Albany, NY oversee the operation of the EUV lithography tool used in the industry-first process to build silicon nanosheet transistors that will enable 5 nanometer (nm) chips. (Photo credit: Dan Corliss)  (Source: SUNY) IBM believes its nanosheet architecture will rank alongside proces technoogy breakthrouths in single-cell DRAMs, chemically amplified photoresists, copper interconnects, silicon-on-insulator, strained materials, multi-core processors, immersion lithography, high-k dielectrics, embedded DRAM, 3D chip stacking and air-gap insulators. Gary Patton, Globalfoundries' chief technology officer and head of worldwide R&D, called the announcement "groundbreaking" and said it demonstrates that Globalfoundries is actively pursuing next-generation technologies at 5nm and beyond.  Also contributing to the Research Alliance's 5nm nanosheets was the SUNY Polytechnic Institute Colleges of Nanoscale Science and Engineering’s NanoTech Complex in Albany, NY.
2017-06-06 00:00 reading:386
  • Week of hot material
  • Material in short supply seckilling
model brand Quote
CD74HC4051QPWRQ1 Texas Instruments
TPS5430DDAR Texas Instruments
TPIC6C595DR Texas Instruments
TL431ACLPR Texas Instruments
TPS61021ADSGR Texas Instruments
PCA9306DCUR Texas Instruments
model brand To snap up
ULQ2003AQDRQ1 Texas Instruments
TXS0104EPWR Texas Instruments
TPS61021ADSGR Texas Instruments
TPS61256YFFR Texas Instruments
TPS5430DDAR Texas Instruments
TPS63050YFFR Texas Instruments
Hot labels
Information leaderboard
  • Week of ranking
  • Month ranking
About us

Qr code of ameya360 official account

Identify TWO-DIMENSIONAL code, you can pay attention to

AMEYA360 mall ( was launched in 2011. Now there are more than 3,500 high-quality suppliers, including 6 million product model data, and more than 1 million component stocks for purchase. Products cover MCU+ memory + power chip +IGBT+MOS tube + op amp + RF Bluetooth + sensor + resistor capacitance inductor + connector and other fields. main business of platform covers spot sales of electronic components, BOM distribution and product supporting materials, providing one-stop purchasing and sales services for our customers.