Do we expect these to lower the prices of memory beyond the usual results of technology decreasing in cost over time? Small gadgets, absolutely - but I am not certain I see how it will change the pricing picture any further than it would normally. #samsungflashmemorychip
If this is both as good as flash, but as fast as DRAM, could we seen an evolution in computing architecture? Imagine, if you will, a world where hard drives and RAM are the same thing. If you wanted more "RAM" for your system, you just partition off larger chunks of your drive to be used.
Not to mention the savings in the board architecture since you wouldn't need the bridge to the RAM and the connection to the hard drives.
Intel tried to push to market a fast, new, proprietary memory format a few years back...it was called RD-RAM...and that was a cluster ef. Why would this be any different than that? Not to mention IA-64, Netburst, etc. #intelphasechangingmemory
@kylecpcs: First, Intel never owned the rights to RDRAM, they licensed it from Rambus who apparently was a pretty horrible technology partner. Their stiff fees and requirements caused Intel and the rest of the industry to eventually dump them.
Second, RDRAM was a *volatile* memory like pretty much every other kind of DRAM. Fast, but you turn the power off and you lose data. This new tech will allow you to have DRAM speeds *and* have the ability to turn the power off. It's not just a competing technology like RDRAM was to SDRAM, it's a memory evolution. Also, this new memory will allow read/write lifetimes on par with SDRAM (or perhaps exceeding it) meaning that all the drawbacks of flash memory are eliminated. Finally it will be possible to truly replace spinning drives with a superior technology. #intelphasechangingmemory
@Lukasz Fabis: Expense doesn't really enter into it. They forced it as a standard (and it was a type of technology, just like DDR) and now I have to deal with clients looking for additional memory for their PC that was only used in half a generation of P4's for about 6 months then died. #intelphasechangingmemory
@kylecpcs:
No, a general type of technology would be something akin to a 32 nm die process, or a particular technique for storing data on optical media, or the difference between DRAM and 1T DRAM.
RDRAM was a proprietary standard of dynamic RAM made by the Rambus Corporation. Netburst was a particular CPU architecture based on the x86 instruction set. IA-64 is a particular instruction set.
We're talking about apples, oranges, and apple smoothies here. Phase-change memory isn't some proprietary rehash of existing standards that's just bound to suck money out of consumer's pockets, and it's not some new competing standard to stir up the status quo. This is a fundamental change in the underlying technology of solid state data storage. Once it becomes viable for mass production, you can bet it will be compatible with existing hardware, in this case, most likely as a USB drive or an SATA SSD.
So yes, expense has everything to do with it. It will likely have a high initial expense, as most things do, and when production scales up and production techniques are refined, prices will reach more accessible levels. #intelphasechangingmemory
All this talk of cramming pixels into a smaller sensor is useless unless the
lens technology can keep up. DSLR lenses are already reaching their limit. My hope is not about shrinking sensors and cramming pixels but for more optically clear lenses that are affordable. Optics aren't crystal clear. They distort, refract, separate light. A pocket cam will always shoot a softer image
than a DSLR, which will shoot softer than a medium format.
@ELPARTO: Yep, that's what it used to look like. You'd run a stong current one way to magnetize the toroids, and... I'd imagine it was by resistance that each one was read... I once saw a memory "core" from an old IBM mainframe that was about the size of a basketball!
So, I wonder what kind of service light a flash image sensor would have, or if the write cycles issue would even apply at all...
It's a zero-sum game. They're shrinking the pixels by 100x, but then their algorithm derives a single super-pixel gray level from 100 pixels.
They're also going to burn a lot of power in their back end processing in order to replace the ADC in conventional CMOS sensors. It's just not true that these burn a lot of power any more.
I strongly suspect that this approach will yield a significantly lower quantum efficiency (QE) than conventional approaches. QE is measure of how many of the photons hitting the sensor get converted in to an electrical signal.
A high QE with high noise is pretty useless, but modern, conventional CMOS sensors have a read noise of just a couple of electrons. That getting very close to high-end CCDs.
You just can't say that CMOS sensors perform poorly in low light anymore. Sure when you make the pixels very small like in cell phones, there's just not enough photons to get a good signal to noise ratio, but large pixel sensors are a different matter.
BTW, this isn't a new idea. People tried this with DRAMs as far back as the 70s.
If they're up to 100x smaller, that should mean you can maintain pixel density and take 100 binary samples per pixel. Using microfilters, I guess you can change the colours and brightness of light that trigger those binary bits, so 100 binary samples per pixel should be roughly capable of producing four 24-bit colour samples.
The question is, how much noise is there, and if you average those 4 colour samples together, do you get a nice pixel with low enough noise?
11/06/09
11/06/09
11/06/09
11/06/09
11/06/09
@mattycakes: Like this...
#samsungflashmemorychip
11/06/09
11/06/09
11/06/09
10/30/09
Not to mention the savings in the board architecture since you wouldn't need the bridge to the RAM and the connection to the hard drives.
Food for thought. #intelphasechangingmemory
10/29/09
Imagine being able to power down your devices and go back to exactly where you left off microseconds after turning them back on.
10/29/09
10/29/09
The difference here is that this is a general type of technology and not one proprietary product.
Hard drives and flash memory were both rather expensive in their respective infancies as well. #intelphasechangingmemory
10/29/09
Second, RDRAM was a *volatile* memory like pretty much every other kind of DRAM. Fast, but you turn the power off and you lose data. This new tech will allow you to have DRAM speeds *and* have the ability to turn the power off. It's not just a competing technology like RDRAM was to SDRAM, it's a memory evolution. Also, this new memory will allow read/write lifetimes on par with SDRAM (or perhaps exceeding it) meaning that all the drawbacks of flash memory are eliminated. Finally it will be possible to truly replace spinning drives with a superior technology. #intelphasechangingmemory
10/29/09
10/29/09
No, a general type of technology would be something akin to a 32 nm die process, or a particular technique for storing data on optical media, or the difference between DRAM and 1T DRAM.
RDRAM was a proprietary standard of dynamic RAM made by the Rambus Corporation. Netburst was a particular CPU architecture based on the x86 instruction set. IA-64 is a particular instruction set.
We're talking about apples, oranges, and apple smoothies here. Phase-change memory isn't some proprietary rehash of existing standards that's just bound to suck money out of consumer's pockets, and it's not some new competing standard to stir up the status quo. This is a fundamental change in the underlying technology of solid state data storage. Once it becomes viable for mass production, you can bet it will be compatible with existing hardware, in this case, most likely as a USB drive or an SATA SSD.
So yes, expense has everything to do with it. It will likely have a high initial expense, as most things do, and when production scales up and production techniques are refined, prices will reach more accessible levels. #intelphasechangingmemory
10/29/09
10/29/09
10/29/09
10/29/09
"Re-route the main computer's Phase-changing Memory Array through the EPS conduits!" Yeah... that'll work. #intelphasechangingmemory
10/30/09
10/13/09
Whoa
10/13/09
10/13/09
10/13/09
10/09/09
lens technology can keep up. DSLR lenses are already reaching their limit. My hope is not about shrinking sensors and cramming pixels but for more optically clear lenses that are affordable. Optics aren't crystal clear. They distort, refract, separate light. A pocket cam will always shoot a softer image
than a DSLR, which will shoot softer than a medium format.
10/12/09
I think we're soon going to reach a point where sensor resolutions are so high no one cares anymore and the optics will become the limiting factor.
Canon and Nikon better start working on cheaper better optics (if they are not already).
10/09/09
10/09/09
10/09/09
So, I wonder what kind of service light a flash image sensor would have, or if the write cycles issue would even apply at all...
10/09/09
They're also going to burn a lot of power in their back end processing in order to replace the ADC in conventional CMOS sensors. It's just not true that these burn a lot of power any more.
I strongly suspect that this approach will yield a significantly lower quantum efficiency (QE) than conventional approaches. QE is measure of how many of the photons hitting the sensor get converted in to an electrical signal.
A high QE with high noise is pretty useless, but modern, conventional CMOS sensors have a read noise of just a couple of electrons. That getting very close to high-end CCDs.
You just can't say that CMOS sensors perform poorly in low light anymore. Sure when you make the pixels very small like in cell phones, there's just not enough photons to get a good signal to noise ratio, but large pixel sensors are a different matter.
BTW, this isn't a new idea. People tried this with DRAMs as far back as the 70s.
10/08/09
The question is, how much noise is there, and if you average those 4 colour samples together, do you get a nice pixel with low enough noise?