Intel and Micron begin production on new breakthrough class of non-volatile memory; 3D Xpoint memory speeds up to 1000 faster than NAND
29 July 2015
Intel Corporation and Micron Technology, Inc. unveiled 3D XPoint technology, a non-volatile memory that has the potential to revolutionize any device, application or service that benefits from fast access to large sets of data. Now in production, 3D XPoint technology is a major breakthrough in memory process technology and the first new memory category since the introduction of NAND flash memory in 1989.
The explosion of connected devices and digital services is generating massive amounts of new data. To make this “big data” useful, it must be stored and analyzed quickly, creating challenges for service providers and system builders who must balance cost, power and performance trade-offs when they design memory and storage solutions. 3D XPoint technology combines the performance, density, power, non-volatility and cost advantages of all available memory technologies on the market today, the partners said. The technology is up to 1,000 times faster and has up to 1,000 times greater endurance than NAND, and is 10 times denser than conventional memory.
For decades, the industry has searched for ways to reduce the lag time between the processor and data to allow much faster analysis. This new class of non-volatile memory achieves this goal and brings game-changing performance to memory and storage solutions.
—Rob Crooke, senior vice president and general manager of Intel’s Non-Volatile Memory Solutions Group
One of the most significant hurdles in modern computing is the time it takes the processor to reach data on long-term storage. This new class of non-volatile memory is a revolutionary technology that allows for quick access to enormous data sets and enables entirely new applications.
—Mark Adams, president of Micron
Following more than a decade of research and development, 3D XPoint technology was built from the ground up to address the need for non-volatile, high-performance, high-endurance and high-capacity storage and memory at an affordable cost. It ushers in a new class of non-volatile memory that significantly reduces latencies, allowing much more data to be stored close to the processor and accessed at speeds previously impossible for non-volatile storage.
The innovative, transistor-less cross point architecture creates a three-dimensional checkerboard where memory cells sit at the intersection of word lines and bit lines, allowing the cells to be addressed individually. As a result, data can be written and read in small sizes, leading to faster and more efficient read/write processes.
3D XPoint innovations include:
Cross Point Array Structure. Perpendicular conductors connect 128 billion densely packed memory cells. Each memory cell stores a single bit of data. This compact structure results in high performance and high density.
Stackable. The initial technology stores 128Gb per die across two stacked memory layers. Future generations of this technology can increase the number of memory layers and/or use traditional lithographic pitch scaling to increase die capacity.
Selector. Memory cells are accessed and written or read by varying the amount of voltage sent to each selector. This eliminates the need for transistors, increasing capacity and reducing cost.
Fast Switching Cell. With a small cell size, fast switching selector, low-latency cross point array, and fast write algorithm, the cell is able to switch states faster than any existing nonvolatile memory technologies today.
 |
3D XPoint technology will sample later this year with select customers, and Intel and Micron are developing individual products based on the technology.
It is progress like this that eventually will enable true autonomous driving. Computing power is still a showstopper for truly self-driving cars. You need to be able to process several ultrahigh definition video streams simultaneously in order to make a self-driving car and you do not want to do it using a lot of energy and space for the required computers. All the systems that collect and process data should not use more than 100 watt combined. We need better computers to do that.
Fortunately, Moore's law will ensure we will get the processing power that are needed. In less than 20 years that law will guarantee that a smart phone like the iPhone 6 will be a 100,000 times more powerful than it currently is without using any more energy. That kind of super computing power is needed in order to run a highly capable artificial intelligence directly on the phone instead of indirectly through a datacenter using the internet as currently is the case for Siri, Cortana and Google now. In my opinion a truly useful self-driving car needs an artificial intelligence that drives the car and understands its passengers spoken language (whatever nationality and dialect) without having to be online using a datacenter. I for one do not want to be on-line when the car is in motion simply because someone with intent could than hack the car and crash it.
In 3 to 5 years I expect we will see the first self-driving taxi services to operate at specific roads that have been cleared to be safe enough to facilitate self-driving vehicles. As the technology improves more roads of greater driving difficulty will be added and eventually the technology will become so god that the cars can drive themselves on any road in any weather condition no matter how poorly that road is maintained with lane lines, traffic signs, road construction, etc.
Posted by: Account Deleted | 29 July 2015 at 02:50 AM
It's not just about processing speed. It's about complexity and liability. You won't see fully autonomous vehicles on the roads in 3-5 years because there are two many edge cases, and drivers can not be relied on to re-focus attention fast enough to take control in a situation too complex for the autonomous system to handle safely.
Volvo is very enthusiastic about the technology, and I applaud their efforts. But their current systems can not even keep the car centered in a lane, it runs along the edge, and will not stay in the lane in a sharp bend.
Autonomous cars will eventually happen. But they will not happen any time soon.
Posted by: electric-car-insider.com | 29 July 2015 at 09:43 AM
E.c.i.
My 3 to 5 year estimate is not based on the Volvo development but on what I am reading and viewing about Google's self-driving car. As you probably know Google is currently testing a fully self driving car in the streets of Mountain View, California and Austin, Texas. Admittedly it only drives on a selected set of roads and it is a neighborhood electric vehicle with a top speed of only 25 miles per hour. They are testing 40 of them now and will expand to 150 in the coming months. They are indeed truly self-driving. No human interaction is needed as a failure in one critical system will be corrected by another back-up system taking over immediately. Google was forced by the authorities to retrofit a gas pedal and a steering wheel but they are entirely not needed. They are not used at all. But they can be used and that is what the bureaucrats needed to hear in order to authorized the test. Google is past testing these vehicles in controlled environments (dummy cities). With 150 vehicles clocking 30,000 miles per year Google will get accident and reliability statics from nearly 5 million miles per year. With about 10 million miles of real world city driving done by the end of 2017 I would expect the next phase is to allow Google to start a taxi service on pre approved roads in 2018. As the cars clock more miles and the software is improved the speed should be raised to 40 miles per hour and so forth until these self-driving taxies are allowed to go anywhere.
See for yourself at http://www.google.com/selfdrivingcar/
and the TED talk at https://www.ted.com/talks/chris_urmson_how_a_driverless_car_sees_the_road
Of cause I expect Tesla to make a big bold move in this area also.
Posted by: Account Deleted | 29 July 2015 at 11:33 AM
@ECI, @Henrik.
It is about complexity and defining the problem.
It does not require several high definition video streams, current resolution is quite enough, though we might want to use Lidar and radar for ranging and seeing in visually cluttered environments (driving by a stand of trees etc.).
Humans are really good at driving with incomplete information, they can drive in fog and snow and heavy rain (when they probably shouldn't). It will be very hard to make computers drive them when the sensors are giving very poor data.
I think it will need a breakthrough in the understanding of how to program a computer to drive a vehicle through busy streets with competing, aggressive traffic.
The only way you will get to this is to have multiple teams trying to do it and building prototypes etc.
You need multiple teams for a proper variety of effort and approaches. It might be a university in Vietnam or Serbia that makes the breakthrough.
Google, for instance think you need cm precision maps, but this has to be wrong as people can drive in cities without maps at all, being able to recognize the road edges from curbs and white lines, and people about to cross etc etc.
(People can drive in strange cities with 2m precision maps in current GPS's).
I don't think liability will be so much of a problem. The cars will be equipped with a battery of cameras and will have video footage of the collision and so it will be easy to assign blame. In a proportion of cases, the cars will be to blame, and they will have to pay out. The government might want to set up a table of costs per injury level.
We already cope with manually driven cars injuring and killing people and for this, we have insurance. The same will apply to auto cars. We might have to write laws that anticipate bugs in systems and limit the payout for deaths to sensible levels (which could be based on current practice).
We expect road deaths to go down (but not to zero) once auto cars are working properly but we'll lose a few along the way as the systems get debugged.
There will be corner cases, and impossible cases (a child running onto a road in front of a car), but as long as silly problems get solved fairly quickly, we should be OK.
You just need legislation to limit the payouts for deaths and injury and standard methods of reading the camera and "black box" data for quick and accurate inquests after crashes.
Posted by: mahonj | 30 July 2015 at 02:23 AM
@mahonj
We completely agree that black box recordings will be able to solve the liability problem by far and large. It is not as easy as it sounds though. Today's cars could equally benefit from black box recordings but practically none cars sold have such boxes and the problem is costs. The 360 degrees cameras, radars, lasers and ultrasound sensors together with the cabling, processors and storage medium could easily add well over 10,000 USD worth of equipment for the black box ability at current prices. Of cause with an autonomous car we only need to add the storage medium perhaps 2 terabytes of SSD drives in a fireproof box will do costing 500 USD and then a backup black box costing another 500 USD.
I would like to stress that Google so far is the only player globally that is testing fully self driving cars on public roads. Everybody else are just testing driver assistance and safety systems which are nice to have but they will not change the world as we know it like the self driving technology that Google is testing. Google's self driving project is therefore by far the most interesting of all the efforts out there on the topic. Google now delivers monthly progress reports on this topic at http://www.google.com/selfdrivingcar/reports/
I would also say it is already a complex effort by multiple groups at hundreds of companies working on different aspects of the self-driving technology from sensors (like Bosh) to processors (like Nvidia) to software (Google) and system integration (every single global auto maker). Google will soon have 150 test cars driving so they must have at least 500 people working on this. Globally I expect almost 10000 engineers working on this up from max 2000 just 5 years ago. Everybody knows and agree that this is a make or break technology for the industry. In a few years consumers will not buy a car unless it is packed with safety and driver assistance systems and in 10 years self driving taxi services could be destroying private ownership of cars at record speed globally changing the world forever.
Posted by: Account Deleted | 30 July 2015 at 04:55 AM
Thanks for sharing the links, Henrik.
It's exciting stuff, and I'm sure that at some point we'll get there. But there a huge gulf between the aspirational goals and present day reality. How many self-parking cars have you personally demoed? It looks really good in the videos. Not nearly as good in real life, today. We'll get there. But it will take a lot of very patient investors. These type of innovations tend to have boom and bust cycles.
Posted by: electric-car-insider.com | 30 July 2015 at 07:48 AM
Anyone who has driven with a partner who drives and concentrates on other things (reads) has experienced the situation where the partner glances up to see a situation (cut off etc) that is alarming without the full historical perspective of the driver. The partner reaction further distresses the driver who is in full control. That has to be similar to a driver who is not paying full attention because he is backup to an auto driving system. The backup driver's reaction time has to be severely compromised due to lack of current/recent situational awareness.
A backup driver seems somewhat useless in most cases. The automatic systems won't 'give up' or alert the driver until it is too late, and they just have to be that much better to start with.
Posted by: william g irwin | 31 July 2015 at 05:09 AM
Improved drivers assistance followed by autonomous driving will be the "in" thing to have on next generation or next decade vehicles, specially on electrified vehicles.
Both will make driving safer, reduce accident, costly damages, injuries and fatalities.
Repair, health care and insurance cost will go down.
Many traffic jams will be reduced and/or avoided.
The soonest we do it the better off everybody will be.
However, since it may increase initial vehicle purchase cost, penny pinching users will object as ususal.
Posted by: HarveyD | 31 July 2015 at 07:17 AM
E.c.i.
I know it takes forever to make just little progress in the auto industry with regard to gassers. The Prius, for instance, only gains 5 mpg in efficiency every 6 years and also handles a little better and gets a little more comfortable every 6 year. The transition to BEVs is also going to be fairly slow. Problem is you need to build 200, 50GWh battery factories to make 100 million long-range BEVs per year and for Tesla to build just one takes 6 years. So we are talking decades for sure before we see a full transition to BEVs.
However, self-driving vehicle technology is just tiny censors, PCs and software. Once you have a fully functioning basic self-driving system that is capable of performing in all types of driving conditions (and it appears Google is getting very close to that goal; the head of this program at Google expect 5 years at most) it is easy to scale up production of the components of that system (the censors, PCs and software). You can go from making 100 of those systems per year to making 100 million of those systems to be sold to automakers in 2 to 3 years. That is impossible with 90kwh battery packs or 200kw electric motors or power controllers. The latter will take decades even if you know how to do it very profitable and you have unrestrained demand. You literally have to transform million of tons of raw materials into these components whereas with sensors and processors we a talking a few 100 tons of raw materials in an industry that is already experienced in making hundreds of millions of copies of these items for the smart phone industry.
The other thing why the transition to self-driving vehicles will be very fast once the first systems are made is that there are 100 of billions of USD to be saved every year and billion of hours of driving time to be saved and literally 1.2 million human lives that are killed every year globally in the traffic because of human errors. They could also be saved along with an even larger number of humans that are permanently crippled every year.
Take it all together, life and livelihood saved, time saved, money saved and the ease of scaling up production and you will understand my optimism for why I think self-driving cars are going to be one of the fastest transition and deployment of new technology that the industrial world have seen so far. Of cause we still need to see that first operational taxi service in action on public roads. My bet is on Google or Tesla to be first with that and they may join forces.
Posted by: Account Deleted | 31 July 2015 at 10:15 AM
@henrik
You could build a decent black box for about $200 in bulk.
You only need 1 or 2 cameras, an accelerometer and some storage (say 16 GB max). You only need to keep about 5 minutes video. You might have an extra camera at the back for rear endings.
It is only a black box for a car, not a plane with 300 people in it. You could stuff it on the lower dash and connect it to the CAN bus for car control signals and power. You might (or might not add a mic). [ You would get a better view from the top of the windscreen, but it would be harder to get power to it ].
Everything you need will already be in an auto car, including the flash ram.
I imagine they are not in cars already because people do not want them, because many accidents are caused by the drivers (rather than external forces) [ I have no figures for this, but it seems reasonable ]
I think a worry for auto cars will be hacking and people taking control of them. The car manufacturers will devise better security protocols, but there will be an army of hackers looking to play with outer people's cars.
Also, we may find that AVs will be very cautious and will drive drivers crazy and cause them to do very risky things. It will be like an army of grannies driving cars.
Posted by: mahonj | 31 July 2015 at 10:34 AM
All good points mahonj.
Henrik, love the optimism, but as a career software developer/system architect/program manager, I know how these things go wrong. Eventually, yes. Soon, no. Especially with the policy/legal issues.
> head of this program at Google expect 5 years at most
There are some projects that if you tell the truth, your project won't get funded (10-20 year projects get DARPA funding, not VC). So everyone lies, even if the managers and even leadership know it's a lie. But it's not called a lie. It's called "an aspirational goal."
Read some of the seminal books on software and system development. Missed schedules, sometimes by many years, are typical. And the more complex the project, the bigger the miss.
It's just the way the real worl works. Ask any engineer.
Posted by: electric-car-insider.com | 31 July 2015 at 12:22 PM
E.c.i.
I hear you and I think you are spot on about lying to superiors to keep things going in many cases but not this case. The case of fuel cell cars definitely qualifies. For the past 40 years lots of people in the auto industry have been saying fuel cell cars are just around the corner and regardless of zero real world results many still seem to believe in it.
Larry Page calls the shots at Google and he does not care anymore about making money short-term. He has all he can spend in a 1000 lifetimes so he will fund the self-driving project regardless of how long it takes. Since college he has been passionate about transportation systems so he want this to happen. I am certain he couldn't care less if Google's boss of self driving cars Chris Umson told him it would probably take 10 years. Chris thinks his team can do it before 2020 and I do not see why he should be lying about that to anyone considering his boss is Larry Page. There is a highly recommendable TED talk about Larry Page and his philosophy for Google and life and about artificial intelligence. He starts to talk about cars 15 min and 15 sec into the 23 min long interview but I recommend viewing it all.
https://www.ted.com/talks/larry_page_where_s_google_going_next#t-1386504
Artificial intelligence is also a big topic in this talk and it is intimately related to self-driving cars in my opinion and also in Google's opinion. Part of Google's self driving software is the system's ability to recognize objects like humans, animals and different types of cars (like an ambulance in action versus an ordinary car) and predict where they go next in order to avoid accident. It is done with programming and learning computer based neural networks about these things.
@mahonj
It does not matter much if it cost 5000 USD or 20000 USD to add self driving technology to a car. Tesla are aiming for a service like for 1 million miles for their BEVs. Add self-driving technology and operate it as a taxi charging 40 cents for 1 million miles doing 100,000 miles per year in 10 years and that taxi will make 400,000 USD. This is plenty to pay for every cost associated with that driving and still make good profits. Today a Taxi cost 1.4 USD per mile but 1 USD is for the taxi driver and overheads that are not paid in a self-driving vehicle. I want to focus on the big picture and not on whether you can get a SSD drive black box for 200 or 500 USD. You are probably right about the 200 USD. However, if I were a company using self-driving taxies I would want a deluxe monitoring system with six, 360 degrees cameras in 4k and radar etc. and a 30 minutes recording to be sure my case in court would be well documented. There will be hordes of people both private and industry lobbyist that will try to coerce money from you or prevent the future from happening because their clients stand to lose from it. So spending extra on that is money well spend.
Posted by: Account Deleted | 31 July 2015 at 02:04 PM
"The transition to BEVs is also going to be fairly slow. Problem is you need to build 200, 50GWh battery factories to make 100 million long-range BEVs per year and for Tesla to build just one takes 6 years. So we are talking decades for sure before we see a full transition to BEVs."
I believe you need to rethink this. A battery factory is a) a building with b) machines in it. It's not the case that there is only one crew in the world that can build building or one factory that can build industrial machines.
Spread them around the world a bit and we could build 200 large buildings in a couple of years. And it probably wouldn't take much longer to pump out the machines to make the batteries. Tesla is likely taking 6 years because they are looking at expected rate of sales growth and assuming that's about right to meet demand.
Posted by: Bob Wallace | 31 July 2015 at 09:10 PM
"It does not matter much if it cost 5000 USD or 20000 USD to add self driving technology to a car."
I doubt it will take anything like that to add self-driving. Digital sensors and lenses are inexpensive. Radar modules are inexpensive. Processing is not all that expensive. Most likely the system will use a series of stills rather than video and do some simple object/distance processing frame to frame to calculate impending problems.
Cars will likely be 'drive by wire'. They somewhat already are. We don't have a lever system operating the carburetor, the computer controls fuel and air flow. We're already driving partially run by wire brakes with our ABS systems. Steering - we might leave the mechanical link in for a generation or two for driver assurance and just stick in a servo motor at the most appropriate point.
Our cars already have basic navigation systems. Maps come with a GPS system that sells for not much more than $50. The car would just need more detail and likely wouldn't store all the maps for the entire country/continent but would download what it needs once we enter the destination. No sense storing details for North Dakota if you're driving from New York to North Carolina.
Black box data? How much would really be needed? At the end of a trip store summary data and any shot sequence during which the car had to perform an out of the ordinary operation (dodge a dog). Compress the heck out of the files and storage required gets very small. Perhaps give the owner of offloading data more than a few months old and then dump it. Just keep the statistical summaries.
Posted by: Bob Wallace | 31 July 2015 at 09:32 PM
Self-driving and insurance -
"Allstate, Geico, State Farm, and others are grappling with innovations that could put a huge dent in their revenue. As carmakers automate more aspects of driving, accidents will likely plunge and car owners will need less coverage. Premiums consumers pay could drop as much as 60 percent in 15 years as self-driving cars hit the roads, says Donald Light, head of the North America property and casualty practice for Celent, a research firm. His message for insurers: “You have to be prepared to see that part of your business shrink, probably considerably.”"
" a system introduced on the 2013 Honda Accord beeps when cars get too close to traffic ahead or leave their lane without signaling. It has had a measurable effect on the frequency of some types of claims: Bodily injury liability losses dropped 40 percent and medical payments decreased 27 percent, according to a 2014 study of insurance claims data"
http://www.bloomberg.com/news/articles/2015-07-30/can-the-insurance-industry-survive-driverless-cars-
Posted by: Bob Wallace | 31 July 2015 at 09:38 PM
Mass manufacturing of high demand goods such as future EV batteries, light carbon fiber wheels, electrified-electronic components and EV vehicles etc, will not be delayed by lack of factories and equipment for very long.
China can already build semi-prefab (in adjacent plant) high rise large hotels in 18 days at the rate of 5 to 6 floors/day, 45+ Km bridges in less than 18 months and 10,000+ Km very high speed e-railroads per year etc.
Building 500 to 1,000 fully equipped flexible adaptive production mega EV battery factories (worldwide) by 2020 or before should not be a major problem. EVs overproduction may be.
Posted by: HarveyD | 01 August 2015 at 09:00 AM
The main cost of automatic cars will be the engineering, which will be a sunk cost. The actual sensors won't have to cost very much if people just use vision and radar (lidar might be a bit more expensive).
There is an Israeli company doing it (for Volvo and others) with a single camera (!)
www.mobileye.com/ also www.magna.com
(However, I wouldn't trust this - it is just too clever).
But I would trust a stereo camera system for motorway driving (with rear cameras for lane change checking).
Also, who says automatic cars have to be electrically driven? they could have ICE or Hybrid engines, they only need electrical control.
A motorway and A-road driving car would have to be able to handle motorways, cloverleafs and (in the UK) roundabouts.
It would also have to be able to manage regions where there are no road markings and very confusing road markings.
(or at least detect these and slow down or hand over control or trust GPS).
Posted by: mahonj | 02 August 2015 at 02:08 AM
@mahonj
Wired has a good piece on the cost of self-driving technology. It cost much more for the level of quality sensors that are needed for fully self-driving cars than just for driver assist systems. Google is so far the only fully self-driving car in testing and its GPS system that is accurate by centimeters cost 6000 USD and its lidar that fires 64 lasers collecting 1 million data points per second cost 60000 USD. Boston consulting does not expect fully self driving technology to cost less than 10,000 USD per car until 2025.
http://www.wired.com/2015/04/cost-of-sensors-autonomous-cars/
However, even a high 20,000 USD per car is trivial to total cost of self-driving cars if it is operated as a taxi service clocking 100,000 miles per year for 10 years making 40 cents per mile or 400,000 USD over its 10 years lifespan.
Posted by: Account Deleted | 02 August 2015 at 04:22 AM
Our unshaved badly dressed city bus drivers cost over $128,500/year each + all the accidents and bad PR they create.
Driveless city buses at $100,000 extra each would be a give away.
It is time for all of them to go!!!!
Posted by: HarveyD | 07 August 2015 at 12:40 PM