To 20nm and beyond: ARM targets Intel with TSMC collaboration

The multi-year deal sees ARM tie itself even closer to TSMC, its chip-fabber of choice, as it looks to capitalise on the company's technology to help it maintain a lead over Intel for chip power efficiency.

building-the-low-power-20nm-ecosystem ARM is ramping up its push to get its highly efficient low-power chips into servers by signing a multi-year agreement with Asian silicon manufacturer TSMC.
Under the deal, the Cambridge-based chip designer has agreed to share technical details with TSMC to help the fabricator make better chips with higher yields, ARM said on Monday. TSMC will also share information, so that ARM can create designs better suited to its manufacturing.

"By working closely with TSMC, we are able to leverage TSMC's ability to quickly ramp volume production of highly integrated SoCs [System-on-a-Chip processors] in advanced silicon process technology," Simon Segars, general manager for ARM's processor and physical IP divisions, said in a statement.

"The ongoing deep collaboration with TSMC provides customers earlier access to FinFET technology to bring high-performance, power-efficient products to market," he added.

The move should keep ARM's chip designs competitive with Intel's in the server market. TSMC's FinFET is akin to Intel's 3D 'tri-gate' method of designing processors with greater densities, which should deliver greater power efficiency and better performance from a cost point of view. 

By tweaking its chips to TSMC's process, ARM chips should deliver good yields on the silicon, keeping prices low while maintaining the higher power efficiency that comes with a lower process node.
ARM's chips dominate the mobile device market, but unlike Intel, it doesn't have a brand presence on the end devices. Instead, companies license its designs, go to a manufacturer, and rebrand the chips under their own name. You may not have heard of ARM, but the Apple, Qualcomm and Nvidia chips in mobile devices, as well as Calxeda and Marvell's server chips, are all based to some degree on based on ARM's low-power RISC-architecture processors.

64-bit processors

As part of the new deal, ARM is expecting to work with TSMC on 64-bit processors. It stressed how the 20nm process nodes provided by the fabber will make its server-targeted chips more efficient, potentially cutting datacentre electricity bills.

"This collaboration brings two industry leaders together earlier than ever before to optimise our FinFET process with ARM's 64-bit processors and physical IP," Cliff Hou, vice president of research and development for TSMC, said in the statement. "We can successfully achieve targets for high speed, low voltage and low leakage."

"We can successfully achieve targets for high speed, low voltage and low leakage" — Cliff Hou, TSMC

However, ARM only released its 64-bit chips in October, putting these at least a year and a half away from production, as licensees tweak designs to fit their devices. Right now, there are few ARM-based efforts pitched at the enterprise, aside from HP's Redstone Server Development platform and a try-before-you-buy ARM-based cloud for the OpenStack software.

Production processes

AMD, like ARM, does not operate its own chip fabrication facilities and so must depend on the facilities of others. AMD uses GlobalFoundries, while ARM licensees have tended to use TSMC. However, both TSMC and GlobalFoundries are a bit behind Intel in terms of the level of detail — the process node — they can make their chips to.
Right now, TSMC is still qualifying its 20nm process for certification by suppliers, while Intel has been shipping its 22nm Ivy Bridge processors for several months. Intel has claimed a product roadmap down to 14nm via use of its tri-gate 3D transistor technology, while TSMC is only saying in the ARM statement it will go beyond 20nm, without giving specifics.

Even with this partnership, Intel looks set to maintain its lead in advanced silicon manufacturing.
"By the time TSMC gets FinFET into production - earliest 2014, it's only just ramping 28nm [now] - Intel will be will into its 2nd generation FinFET buildout," Malcolm Penn, chief executive of semiconductor analysts Future Horizons, told ZDNet. This puts Intel "at least three years ahead of TSMC. Global Foundries will be even later."

Intel has noticed ARM's rise and has begun producing its own low-power server chips under the Centerton codename. However, these chips consume 6W compared with ARM's 5W.
At the time of writing, neither ARM nor TSMC had responded to requests for further information. Financial terms, if any, were not disclosed.

Editing your FPGA source

I noted that in a recent poll of FPGA developers, emacs was far and away the most popular VHDL and Verilog editor. There are a few reasons for this – namely, emacs comes with packages for editing your HDL of choice. For those of us not wanting to install (and learn) the emacs operating system, I got Notepad++ to work with these packages.

Notepad++ already has VHDL and Verilog highlighting along with other advanced text editor features, but I wanted templates, automated declarations and beautification. To do this, he used the FingerText to store code as snippets and call them up at the wave of a finger.

As I writes his code, the component declarations constantly need to be updated, and with the help of a Perl script I can update them with the click of a hotkey. Beautification is a harder nut to crack, as Notepad++ doesn’t even have a VHDL or Verilog beautifier plugin. This was accomplished by installing emacs and running the beautification process as a batch script. Nobody can have it all, but we’re thinking that this method of getting away from emacs is pretty neat.

VLSID 2013 - 26th International Conference on VLSI Design 2013

vlsi_2013Venue: Hyatt Regency, Pune, India

This joint conference is a forum for researchers and designers to present and discuss current topics in VLSI design, electronic design automation, embedded systems, and enabling technologies. Two days of tutorials will be followed by three days of regular paper sessions, special sessions, and embedded tutorials. Industry presentation sessions along with exhibits, panel discussions, Design Contest, and Education Forum round off the program.

The theme for the conference is Green Technology - A New Era for Electronics, which explores the ability of VLSI and embedded circuits and systems to positively impact the environment. Example areas under the theme include (but are not restricted to) designing energy-efficient VLSI circuits, improving the efficiency of energy-hungry applications such as data centers, developing intelligent monitoring and control systems such as smart grids, and using integrated circuits or embedded systems to leverage novel green technologies.

Pune city, also known as the Oxford of the East, has witnessed huge growth in its VLSI and embedded community over the past few years and is proud to host its very first VLSI Design Conference. The conference organizing committee is looking forward to making this an unforgettable experience for all attendees.

Conference Program:

Day 1


5th January 2013



Day 2


6th January 2013



Day 3


7th January 2013


Inaugural, Technical Sessions, Student

Day 4


8th January 2013


Valedictory/Award, Technical Sessions, Student Conference

Day 5


9th January 2013


Technical Sessions, RASDAT 2013 workshop

Day 6


10th January 2013


RASDAT 2013 workshop

Proposal submission links now available!
Papers, Tutorials, User/designer track submissions, Design contest and Embedded Tutorials, Special Sessions, Panels.
Deadline for Regular Paper submission has been extended to 24th July, 2012. Please submit Tutorial, User/designer track submissions, Design contest and Embedded Tutorials, Special sessions, Panels proposals by 2nd August, 2012.

Important Dates:

Paper Submissions


24th July, 2012

Tutorial Submissions


2nd August, 2012

User/Designer Submissions


2nd August, 2012

Call for embedded tutorials, special sessions, and panels


2nd August, 2012

Design Contest Submission


15th September, 2012

Acceptance of notification


7th September, 2012

Camera ready paper due


1st October, 2012


MIT engineers Innovates an “Intelligent co-pilot” for cars

IT engineers have developed a semi-autonomous vehicle safety system that takes over if the driver does something stupid.

The system uses an onboard camera and laser rangefinder to identify hazards. An algorithm analyzes the data and identifies safe zones — avoiding, for example, barrels in a field, or other cars on a roadway.
The driver's in charge - until the system recognizes that the vehicle's about to exit a safe zone and takes over.

"The real innovation is enabling the car to share [control] with you," says PhD student Sterling Anderson. "If you want to drive, it’ll just make sure you don’t hit anything."

The team's approach is based on identifying safe zones, or 'homotopies', rather than specific paths of travel. Instead of mapping out individual paths along a roadway, the researchers divide a vehicle’s environment into triangles, with certain 'constrained' triangle edges representing an obstacle or a lane’s boundary.

If a driver looks like crossing a constrained edge — for instance, if he’s fallen asleep at the wheel and is about to run into a barrier  — the system takes over, steering the car back into the safe zone.

The system works well in tests, say its designers: in more than 1,200 trials of the system, with , there have only been a few collisions, mostly when glitches in the vehicle’s camera failed to identify an obstacle.

One possible problem with the system, though, is that it could give drivers a false sense of confidence on their own abilities.

Using it, says Anderson, "You’d say, ‘Hey, I pulled this off,’ and you wouldn’t know that the car is changing things behind the scenes to make sure the vehicle remains safe, even if your inputs are not."

He and Iagnemma are now exploring ways to tailor the system to various levels of driving experience.

They're also hoping to pare down the system to identify obstacles using a single cellphone.

"You could stick your cellphone on the dashboard, and it would use the camera, accelerometers and gyro to provide the feedback needed by the system," says Anderson.

"I think we'll find better ways of doing it that will be simpler, cheaper and allow more users access to the technology."

Difference between RDIMM and UDIMM

There are some differences between UDIMMs and RDIMMs that are important in choosing the best options for memory performance. First, let’s talk about the differences between them.

RDIMMs have a register on-board the DIMM (hence the name “registered” DIMM). The register/PLL is used to buffer the address and control lines and clocks only. Consequently, none of the data goes through the register /PLL on an RDIMM (PLL is Phase Locked Loop. On prior generations (DDR2), the Register - for buffer the address and control lines - and the PLL for generating extra copies of the clock were separate, but for DDR3 they are in a single part).

There is about a one clock cycle delay through the register which means that with only one DIMM per channel, UDIMMs will have slightly less latency (better bandwidth). But when you go to 2 DIMMs per memory channel, due to the high electrical loading on the address and control lines, the memory controller use something called a “2T” or “2N” timing for UDIMMs.

Consequently every command that normally takes a single clock cycle is stretched to two clock cycles to allow for settling time. Therefore, for two or more DIMMs per channel, RDIMMs will have lower latency and better bandwidth than UDIMMs.

Based on guidance from Intel and internal testing, RDIMMs have better bandwidth when using more than one DIMM per memory channel (recall that Nehalem has up to 3 memory channels per socket). But, based on results from Intel, for a single DIMM per channel, UDIMMs produce approximately 0.5% better memory bandwidth than RDIMMs for the same processor frequency and memory frequency (and rank). For two DIMMs per channel, RDIMMs are about 8.7% faster than UDIMMs.

For the same capacity, RDIMMs will be require about 0.5 to 1.0W per DIMM more power due to the Register/PLL power. The reduction in memory controller power to drive the DIMMs on the channel is small in comparison to the RDIMM Register/PLL power adder.

RDIMMs also provide an extra measure of RAS. They provide address/control parity detection at the Register/PLL such that if an address or control signal has an issue, the RDIMM will detect it and send a parity error signal back to the memory controller. It does not prevent data corruption on a write, but the system will know that it has occurred, whereas on UDIMMs, the same address/control issue would not be caught (at least not when the corruption occurs).

Another difference is that server UDIMMs support only x8 wide DRAMs, whereas RDIMMs can use x8 or x4 wide DRAMs. Using x4 DRAMs allows the system to correct all possible DRAM device errors (SDDC, or “Chip Kill”), which is not possible with x8 DRAMs unless channels are run in Lockstep mode (huge loss in bandwidth and capacity on Nehalem). So if SDDC is important, x4 RDIMMs are the way to go.

In addition, please note that UDIMMs are limited to 2 DIMMs per channel so RDIMMs must be used if greater than 2 DIMMs per channel (some of Dell’s servers will have 3 DIMMs per channel capability).
In summary the comparison between UDIMMs and RDIMMs is

  • Typically UDIMMs are a bit cheaper than RDIMMs
  • For one DIMM per memory channel UDIMMs have slightly better memory bandwidth than RDIMMs (0.5%)
  • For two DIMMs per memory channel RDIMMs have better memory bandwidth (8.7%) than UDIMMs
  • For the same capacity, RDIMMs will be require about 0.5 to 1.0W per DIMM than UDIMMs
  • RDIMMs also provide an extra measure of RAS
    • Address / control signal parity detection
    • RDIMMs can use x4 DRAMs so SDDC can correct all DRAM device errors even in independent channel mode
  • UDIMMs are currently limited to 1GB and 2GB DIMM sizes from Dell
  • UDIMMs are limited to two DIMMs per memory channel


DIMM Count and Memory Configurations

Recall that you are allowed up to 3 DIMMs per memory channel (i.e. 3 banks) per socket (a total of 9 DIMMs per socket). With Nehalem the actually memory speed depends upon the speed of the DIMM itself, the number of DIMMs in each channel, the CPU speed itself. Here are some simple rules for determining DIMM speed.

  • If you put only 1 DIMM in each memory channel you can run the DIMMs at 1333 MHz (maximum speed). This assumes that the processor supports 1333 MHz (currently, the 2.66 GHz, 2.80 GHz, and 2.93 GHz processors support 1333 MHz memory) and the memory is capable of 1333 MHz
  • As soon as you put one more DIMM in any memory channel (two DIMMs in that memory channel) on any socket, the speed of the memory drops to 1066 MHz (basically the memory runs at the fastest common speed for all DIMMs)
  • As soon as you put more than two DIMMs in any one memory channel, the speed of all the memory drops to 800 MHz

So as you add more DIMMs to any memory channel, the memory speed drops. This is due to the electrical loading of the DRAMs that reduces timing margin, not power constraints.
If you don’t completely fill all memory channels there is a reduction in the memory bandwidth performance. Think of these configurations as “unbalanced” configurations from a memory perspective.

Key Features of upcoming DDR4 memory

Can you believe DDR3 has been present in home PC systems for three years already? It still has another two years as king of the hill before DDR4 will be introduced, and the industry currently isn’t expecting volume shipments of DDR4 until 2013.

JEDEC isn’t due to confirm the DDR4 standard until next year, but following on from the MemCon Tokyo 2010, Japanese website PC Watch has combined the roadmaps of several memory companies on what they expect DDR4 to offer.

Following are the new features proposed for DDR4:

  • Three data width offerings: x4, x8 and x16
  • New JEDEC POD12 interface standard for DDR4 (1.2V)
  • Differential signaling for the clock and strobes
  • New termination scheme versus prior DDR versions: In DDR4, the DQ bus shifts termination to VDDQ, which should remain stable even if the VDD voltage is reduced over time.
  • Nominal and dynamic ODT: Improvements to the ODT protocol and a new Park Mode allow for a nominal termination and dynamic write termination without having to drive the ODT pin
  • Burst length of 8 and burst chop of 4
  • Data masking
  • DBI: to help reduce power consumption and improve data signal integrity, this feature informs the DRAM as to whether the true or inverted data should be stored
  • New CRC for data bus: Enabling error detection capability for data transfers – especially beneficial during write operations and in non-ECC memory applications.
  • New CA parity for command/address bus: Providing a low-cost method of verifying the integrity of command and address transfers over a link, for all operations.
  • DLL off mode supported

It looks like we should expect frequencies introduced at 2,133MHz and it will scale to over 4.2GHz with DDR4. 1,600MHz (10ns) could still well be the base spec for sever DIMMs that require reliability, but it’s expected that JEDEC will create new standard DDR3 frequency specifications all the way up to 2,133MHz, which is where DDR4 should jump off.

As the prefetch per clock should extend to 16 bits (up from 8 bits in DDR3), this means the internal cell frequency only has to scale the same as DDR2 and DDR3 in order to achieve the 4+GHz target.

The downside of frequency scaling is that voltage isn’t dropping fast enough and the power consumption is increasing relative to PC-133. DDR4 at 4.2GHz and 1.2V actually uses 4x the power of SDRAM at 133MHz at 3.3V. 1.1V and 1.05V are currently being discussed, which brings the power down to just over 3x, but it depends on the quality of future manufacturing nodes – an unknown factor.

While 4.2GHz at 1.2V might require 4x the power it’s also a 2.75x drop in voltage for a 32 fold increase in frequency: that seems like a very worthy trade off to us – put that against the evolution of power use in graphics cards for a comparison and it looks very favorable.


One area where this design might cause problems is enterprise computing. If you’re using a lot of DIMMs, considerably higher power, higher heat and higher cost aren’t exactly attractive. It’s unlikely that DDR4 4.2GHz will reach a server rack near you though: remember most servers today are only using 1,066MHz DDR3 whereas enthusiast PC memory now exceeds twice that.

Server technology will be slightly different and use high performance digital switches to add additional DIMM slots per channel (much like PCI-Express switches we expect, but with some form of error prevention), and we expect it to be used with the latest buffered LR-DIMM technology as well, although the underlying DDR4 topology will remain the same.

This is the same process the PCI bus went through in its transition to PCI-Express: replacing anything parallel nature with a serial approach. DDR4 will become a point-to-point bus and the parallelism is being left with the memory controller itself with multiple memory channels.

If we look at Intel’s upcoming LGA2011 socket that is anticipated to use a quad-channel memory interface and a single DIMM per channel, it’s now quite obvious that future CPUs using this socket stand a good chance of using DDR4, especially as LGA1366 has had a well defined three year lifespan. In the same timeframe DDR4 could see considerable market acceptance so it’s a smart move by Intel.

The big questions remain to be answered then: is it a consumer (cost) friendly process and how well does TSV cope with overclocking? We’ll have to wait for the first samples in 2011-2012 to find out.

The Story Of Electronics

Here is a great video - The Story of Electronics by The Story of Stuff Project. It outlines how electronic manufacturers are "Designing for the dump".
As consumers we must pressure these companies to take ownership of electronics from cradle to grave, and change from "dump designers" to sustainable designers.
This is where Design For Sustainability Software (DFS) can be integral to changing the disposable mindset of these organizations.

Designed for the Dump
Many electronic products are designed for the dump. They have short-life spans, or become obsolete quickly. They are often expensive to repair, and sometimes it’s difficult to find parts. Many consumer-grade electronics products are cheaper to replace than to fix even if you can find someone to fix it.  Because they are designed using  many hazardous compounds, recycling these products involves processing toxic material streams, which is never 100% safe.

Some of the problematic toxic materials that must be removed before recycling are lead in cathode ray tube (CRT) TV monitors and mercury lamps in LCD screens, as well as PVC, flame retardants, and other toxic additives in plastic components.

Designed to Last
Before electronics companies can make the claim that they are green and sustainable, they must shift away from producing “disposable” products designed with a limited lifespan (planned obsolescence) and towards products that are designed to last. Instead of purchasing products with high failure rates and the need for frequent replacement, we should be able to choose long-living, upgradeable goods that have long warranties and can be efficiently repaired and recycled.

Difference between DDR3 and DDR4 RAM

DDR4 is the next evolution in DRAM, bringing even higher performance and more robust control features while improving energy economy for enterprise, micro-server, tablet, and ultrathin client applications. The following table compares some of the key feature differences between DDR3 and DDR4.




  DDR4 Advantage

Voltage (core and I/O)



Reduces memory power demand

VREF inputs

2 – DQs and CMD/ADDR


VREFDQ now internal

Low voltage standard


(DDR3L at 1.35V)


(likely 1.05V)

Memory power reductions

Data rate (Mb/s)

800, 1066, 1333, 1600, 1866, 2133

1600, 1866, 2133, 2400, 2667, 3200

Migration to higher‐speed I/O




Better enablement for large-capacity memory subsystems

Internal banks



More banks

Bank groups (BG)



Faster burst accesses

tCK – DLL enabled

300 MHz to 800 MHz

667 MHz to 1.6 GHz

Higher data rates

tCK – DLL disabled

10 MHz to 125 MHz (optional)

Undefined to 125 MHz

DLL-off now fully supported

Read latency



Expanded values

Write latency



Expanded values

DQ driver (ALT)



Optimized for PtP (point-to-point) applications

DQ bus



Mitigate I/O noise and power

RTT values (in Ω)

120, 60, 40, 30, 20

240, 120, 80, 60, 48, 40, 34

Support higher data rates

RTT not allowed

READ bursts

Disables during READ bursts


ODT modes

Nominal, dynamic

Nominal, dynamic, park

Additional control mode; supports OTF value change

ODT control

ODT signaling required

ODT signaling not required

Ease of ODT control, allows non-ODT routing on PtP applications

Multipurpose register (MPR)

Four registers – 1 defined, 3 RFU

Four registers – 3 defined, 1 RFU

Provides additional specialty readout

Popular Posts