It’s increasingly rare for mechanical drives to make headlines, but Seagate is beating the PR drum over an advancement that promises to drastically increase the capacity of hard drives. The company has achieved a storage density of 1 terabit per square inch, about 55% more than today’s 620 gigabits per square inch. More abstractly, Seagate says that’s more bits per square inch than our Milky Way galaxy has stars, which astronomers estimate between 200 and 400 billion.
At 620Gb per square inch, current 3.5-inch HDDs peak at 3TB, while Seagate’s 2.5-inch consumer drives max out at 750GB. The new tech will roughly double that to 6TB and 2TB when it arrives “later this decade” and it will lead to astronomical capacities of up to 60TB over the following 10 years. Seagate hit the milestone with heat-assisted magnetic recording (HAMR), which the company hails as a next-gen successor to 2006’s perpendicular magnetic recording (PMR).
PMR is expected to peak at approximately 1Tb per square inch in the next few years, which is essentially the starting density of HAMR drives. “The technology offers a scale of capacity growth never before possible, with a theoretical areal density limit ranging from 5 to 10 terabits per square inch — 30TB to 60TB for 3.5-inch drives and 10TB to 20TB for 2.5-inch drives,” Seagate explained in its press release.
Along with avoiding concrete launch windows, the company omitted details about how HAMR works. We assume that’s mostly because your eyes would glaze over, and not for competitive reasons as Fujitsu (acquired by Toshiba in 2009), Hitachi and presumably others have been tinkering with HAMR and other technologies for many years. A 2006 article by CNET does a good job of breaking things down.
The gist of it: Data is stored in bits that contain hundreds of cobalt-platinum grains representing either a 1 or 0. To increase capacities, engineers shrink the size of bits and grains. However, with current tech, we’re approaching a point where further shrinkage could cause grains to flip between 1 and 0 at room temperature, resulting in data corruption. Reducing the number of grains per bit presents other issues.
CNET reported at the time that Seagate and others planned to solve the problem by replacing cobalt-platinum grains with iron-platinum ones that wouldn’t flip at room temperature. Thus enters the “heat-assisted” part of HAMR: drives have an integrated laser to heat the bits and record data. It’s safe to assume other hard drive makers won’t take Seagate’s announcement lying down, so it’ll be interesting to see whether their next-gen drives use similar HAMR-based solutions or something entirely different.
**HDD image via Vitaly Korovin/Shutterstock
IBM News room – 2012-03-08 Made in IBM Labs: Holey Optochip First to Transfer One Trillion Bits of Information per Second Using the Power of Light – United States.
LOS ANGELES – 08 Mar 2012: IBM (NYSE: IBM) scientists today will report on a prototype optical chipset, dubbed “Holey Optochip”, that is the first parallel optical transceiver to transfer one trillion bits – one terabit – of information per second, the equivalent of downloading 500 high definition movies. The report will be presented at the Optical Fiber Communication Conferencetaking place in Los Angeles.
With the ability to move information at blazing speeds – eight times faster than parallel optical components available today – the breakthrough could transform how data is accessed, shared and used for a new era of communications, computing and entertainment. The raw speed of one transceiver is equivalent to the bandwidth consumed by 100,000 users at today’s typical 10 Mb/s high-speed internet access. Or, it would take just around an hour to transfer the entire U.S. Library of Congress web archive through the transceiver.
Progress in optical communications is being driven by an explosion of new applications and services as the amount of data being created and transmitted over corporate and consumer networks continues to grow. At one terabit per second, IBM’s latest advance in optical chip technology provides unprecedented amounts of bandwidth that could one day ship loads of data such as posts to social media sites, digital pictures and videos posted online, sensors used to gather climate information, and transaction records of online purchases.
“Reaching the one trillion bit per second mark with the Holey Optochip marks IBM’s latest milestone to develop chip-scale transceivers that can handle the volume of traffic in the era of big data,” said IBM Researcher Clint Schow, part of the team that built the prototype. “We have been actively pursuing higher levels of integration, power efficiency and performance for all the optical components through packaging and circuit innovations. We aim to improve on the technology for commercialization in the next decade with the collaboration of manufacturing partners.”
Optical networking offers the potential to significantly improve data transfer rates by speeding the flow of data using light pulses, instead of sending electrons over wires. Because of this, researchers have been looking for ways to make use of optical signals within standard low-cost, high-volume chip manufacturing techniques for widespread use.
Photomicrograph of IBM Holey Optochip. Original chip dimensions are 5.2 mm x 5 .8 mm.
Using a novel approach, scientists in IBM labs developed the Holey Optochip by fabricating 48 holes through a standard silicon CMOS chip. The holes allow optical access through the back of the chip to 24 receiver and 24 transmitter channels to produce an ultra-compact, high-performing and power-efficient optical module capable of record setting data transfer rates.
The compactness and capacity of optical communication has become indispensable in the design of large data-handling systems. With that in mind, the Holey Optochip module is constructed with components that are commercially available today, providing the possibility to manufacture at economies of scale.
Consistent with green computing initiatives, the Holey Optochip achieves record speed at a power efficiency (the amount of power required to transmit a bit of information) that is among the best ever reported. The transceiver consumes less than five watts; the power consumed by a 100W light bulb could power 20 transceivers. This progress in power efficient interconnects is necessary to allow companies who adopt high-performance computing to manage their energy load while performing powerful applications such as analytics, data modeling and forecasting.
By demonstrating unparalleled levels of performance, the Holey Optochip illustrates that high-speed, low-power interconnects are feasible in the near term and optical is the only transmission medium that can stay ahead of the accelerating global demand for broadband. The future of computing will rely heavily on optical chip technology to facilitate the growth of big data and cloud computing and the drive for next-generation data center applications.
Technical Aspects of the Holey Optochip
Photomicrograph of the back of the IBM Holey Optochip with lasers and photodectors visible through substrate holes.
Parallel optics is a fiber optic technology primarily targeted for high-data, short-reach multimode fiber systems that are typically less than 150 meters. Parallel optics differs from traditional duplex fiber optic serial communication in that data is simultaneously transmitted and received over multiple optical fibers.
A single 90-nanometer IBM CMOS transceiver IC with 24 receiver and 24 transmitter circuits becomes a Holey Optochip with the fabrication of forty-eight through-silicon holes, or “optical vias” – one for each transmitter and receiver channel. Simple post-processing on completed CMOS wafers with all devices and standard wiring levels results in an entire wafer populated with Holey Optochips. The transceiver chip measures only 5.2 mm x 5.8 mm. Twenty-four channel, industry-standard 850-nm VCSEL (vertical cavity surface emitting laser) and photodiode arrays are directly flip-chip soldered to the Optochip. This direct packaging produces high-performance, chip-scale optical engines. The Holey Optochips are designed for direct coupling to a standard 48-channel multimode fiber array through an efficient microlens optical system that can be assembled with conventional high-volume packaging tools.
Usable data rate is 400 Gbps, 4x current top speeds; should allow for supercharged smartphone networksThere are two major determinants of cellular data network speeds. The first is the physical broadcast infrastructure, which takes into account factors such as number of towers, placement, type of spectrum, and amount of spectrum available. The second major determinant is the physical network backbone — typically fiber cable.
Every signal that goes to or from your smartphone must be transmitted through a fiber backbone. The faster that backbone is, the faster a carrier’s services become, regardless of wireless transmission technology.
The “T-Labs” research group of T-Mobile USA’s German parent Deutsche Telekom AG (ETR:DTE) made a splash this week, announcing [press release] that it had worked the kinks out of ultra fast fiber optic transmission, which travel at a theoretical data transfer speed of 512 Gbps (or “the simultaneous transmission of 77 music CDs” as T-Labs puts it).
Fiber optic communications just got much faster. [Image Source: AllPosters]
Real world performance isn’t far behind. Deutsche Telekom observed real world speeds of 400 Gbps during a 734 km round-trip along a single-optical fiber test channel running between Hanover and the capital city of Berlin.
400 Gbps is a pretty impressive figure, given that the current fastest deployed fiber networks run at around 100 Gbps, with most networks well behind that even. T-Labs plans to bundle together 48 of the single channels into a bundle that will offer a combined throughput of around 18.75 Tbps (18,750,000,000,000 bit/s) (24.6 Tbps, theoretical).
Deutsche Telekom describes the breakthrough in terms that surely would rile the Recording Industry Association of America — “A collection of 3,696 CDs could thus be transferred over a single optical fiber — a strand thinner than a human hair — at the same time.”
The cutting edge research is part of the company’s “Optically Supported IP Router Interfaces” (OSIRIS) project.
“I was working in the lab, late one night…” [Image Source: T-Labs]
For the professional network and electrical engineers T-Mobile offers up some juicy technical tidbits:
The Telekom OSIRIS (Optically Supported IP Router Interfaces) research project realized transmission at a speed of 512 Gbit/s (400 Gbit/s usable bit rate) on a 100 GHz wavelength channel over a distance of 734 km, thus demonstrating a spectral sensitivity of 5 bits/s/Hz in the Deutsche Telekom network.
This tremendous transmission performance was reached using innovative transmission technology with two carrier frequencies, two polarization planes, 16-QAM quadrature amplitude modulation and digital offline signal processing for the equalization of fiber influences with soft-FEC forward error correction decoding in the receiver.The WDM transmission link consisted of a total of 14 standard single-mode fiber sections with dispersion compensation as required for the neighboring conventional 10 Gbit/s channels. The high optical input powers of the conventional 10 Gbit/s channels and the dispersion compensation in the fiber sections interfere with the innovative transmission technology due to nonlinearities, including self-phase modulation by the higher input power and cross-phase modulation by the adjacent channels. Despite these worst-case conditions, it was possible to demonstrate the transfer of the innovative high-speed signal simultaneously with conventional 10 Gbit/s signals in adjacent channels in an existing system.
So what’s the best part about the new technology? It requires no fiber replacement — it merely requires new receivers/transmitters. Deutsche Telekom’s T-Labs believes that it is read for commercialization, once the technology is incorporated into the terminal equipment from network vendors like Sweden’s Ericsson SpA (STO:ERIC B), China’s Huawei, France’s Alcatel-Lucent (EPA:ALU), and Nokia Siemens (a joint venture between Finland’s Nokia Oyj. (HEL:NOK1V) and Germany’s Siemens AG (ETR:SIE)).
AMD’s heavily threaded Bulldozer, APUs, GCN are good fits for Seamicro’s compact cloud computing serversAdvanced Micro Devices, Inc. (AMD) has struggled mightily in the server market in recent years, seeing its market share fall from nearly 15 percent in 2007 to less than half that — roughly 6.5 percent in 2011.
I. AMD Server Division — In Need of a Turnaround
AMD can try to write off part of its struggles to rival Intel Corp. (INTC) using anti-competitive techniques to squelch its performance during its strong years in the middle of the last decade, a big part of the troubles have come due to AMD’s trailing die shrink timing, which has not improved since it spun off its fabs. While AMD finally dropped a new architecture (Bulldozer) in Sept. 2011, it disappointed in clock speeds and power performance — something that may be attributable to die shrinks. Difficulty getting to 32 nm may have left AMD with too little time to thoroughly test and refine the new cores.
Approximately 21.89 percent of AMD’s market share is tied up in its server sales, so clearly this is a major issue for the company and its shareholders. AMD desperately needed a new tactic. While allowing competitive interplay between Taiwan Semiconductor Manufacturing Comp., Ltd. (TPE:2330) and GlobalFoundries in die shrinks may be a potential long term solution, AMD needed something more immediate.
That’s why the news of its acquisition of SeaMicro for $334M USD (a mix of $281M USD cash and stock) is a bit surprising, but a bit unsurprising. The small 80-person Silicon Valley server maker is known as a premium maker of highly dense and power-efficient servers. It sells heavily to large-scale cloud computing businesses.
AMD’s stock price is heavily dependent on server performance. [Image Source: Trefis]
II. Meet SeaMicro
The move is also a boon to Santa Clara, California manufacturer NBS. Unlike Apple, Inc. (AAPL), Hewlett-Packard Comp. (HPQ), Dell, Inc. (DELL) and others, SeaMicro doesn’t have its servers assembled by Chinese laborers working under sweatshop like conditions. It’s made in America, by blue collar workers earning a respectable living.
While it only spends a tenth of the research and development budget (~$50M USD per year) as Dell or HP, SeaMicro’s product is viewed as very competitive from a technology basis. But SeaMicro can work intimately with its American manufacturing partner, building prototypes, trialing optimizations, and working out bugs before production hits.
SeaMicro makes its servers in California — not China. It contracts NBS, a small local manufacturer (pictured). [Image Source: Ariel Zambelich/Wired]
All of this is good news for AMD; as SeaMicro’s strength in terms of power and density could offset its weaknesses in power performance, while accentuate its strengths in highly-threaded performance.
SeaMicro currently exclusively sells Intel-based servers — a mixture of Xeon (Sandy Bridge) based tightly-packed 10 RU designs and mixed 10 RU designs incorporating Intel Atom chips for lighter workloads. The Atom servers use the dual-core 64-bit Atom N570 chip (8.5W TDP). SeaMicro’s unique 10 RU form factor squeezes one to two tower racks into a single compact box-like form factor.
SeaMicro makes compact 10 RU “box” servers. [Image Source: SeaMicro]
AMD pledges — for now — to continue to make Intel-based SeaMicro servers. But it states that special AMD Opteron-based designs will be released before the end of the year.
SeaMicro claims four-fold power reduction and six-fold space reduction by eliminating the typical busy server chipset to just three chips, via proprietary interconnect technology.
SeaMicro’s server boards drastically slash space and power via custom chipsets.
[Image Source: SeaMicro]
The approach is rather different from the more traditional designs of SeaMicro’s primary competitors.
III. Folding in Piledriver, APUs, GCN GPUs, ARM into Thread-Shredding Beasts
In the long term this deal makes a lot of sense. AMD, given its scant stake (and given SeaMicro’s modest market share) can likely phase out Intel’s designs. In cloud workloads Bulldozer and its successor Piledriver could truly shine in the one area AMD currently beats Intel — thread performance.
Look at AMD’s Feb. 2012 roadmap, there were hints at the pending acquisition, which SeaMicro CEO Andrew Feldman says happened “unbelievably quickly.”
AMD could also drop in Hondo (ultra-low power) or Brazos 2.0 (low power) cores in 2012. Then in 2013 it can follow with Temash (ultra-low power) and Kabini (low power).
The Brazos C-50 and C-60 chips already are on par with the N570 in a dual-cores performing at 9W, though they lag in clock speed (1.0 GHz). But recall, that these chips have a beefier GPU than Atom.
One possibility is that AMD may deliver variants of its low power cores without the GPU. Alternatively, once it can incorporate its new compute-friendly Graphics Core Next GPU architecture, it could use on-APU GPU computing to handle cloud workloads.
Assuming AMD explores these tracks thoroughly, its new subsidiary could soon be producing some sweet thread-shredding mixed designs.