I thought this was going to mean each stack was able to directly talk to the controller, since all stacks are resting on an interposer thing. But actually there is still a logic controller slice at the bottom of the stack, not at a right angle to the stack.
Instead of HBM microbumps between layers there is a more compact/dense TSV ("fusion bonded via-in-one") system. Intel once more showing their strong chiplet packing prowess! The claim is that thermals are still much better somehow, in spite of volumetric cell density increasing (from thinner layers). The demo has 8+1 dram+controller layers.
A search for patents by Stephen Morein who is listed as the CTO of SAIMEMORY shows
this one, assigned to Intel, and published strangely close to the press release;
That slide is wrong, it's stacked the same way HBM is, the key difference is that they don't use microbumps anymore and instead have a hybrid bond system using copper to copper bonding. Copper is a good thermal conductor and the via is going through the entire wafer which in turn means that they are forming an uninterrupted copper cylinder that goes through all the layers. It's as if they put heat pipes inside the chip.
Apologies, yeah, that slide seems inaccurate now. They also discuss magnetically coupled wireless interface to the host, lol, yeah, no.
The newer slides show a more conventionally stacked die arrangement. With yet a third way of connecting stacks, that at least visually looks different than via-in-one columns and more like tabs between the layers, and definitely not wireless. Hard to guess what to expect from this assorted material.
That song is etched into my memory and bring back a flood of great feelings. Daily Motion is giving me a playback error, but here's a copy from youtube: https://www.youtube.com/watch?v=xb_PyKuI7II
I would love to be in the room when things like this are pitched. First, how high was the person that came up with the idea? Did the rest of the people get contact highs from how high this person clearly must have been to run with it? Also, what did the voice actors think? "Wait, this isn't for kids? Okay, it's your money"
Not that crazy. It's about the most basic structure you can make. Hard to make a better wheel.
The closest thing I can think of that's come close to maybe challenging DRAM is HP's memristors but those really didn't pan out (probably too much power consumption).
Pet peeve: stupid analogy seeing how wheels kept being improved throughout the millennia with every new technology. The only thing in common is that it's round.
Similarly, DRAM in any way you see it has been improving to the point of barely being recognizable since the 70s.
That said, DIMMs and the whole bus idea is in dire need of getting a new type of bearing.
> That said, DIMMs and the whole bus idea is in dire need of getting a new type of bearing.
IBM has been using their own memory bus technology for both their POWER and Z machines. IIRC, it’s somewhat reminiscent of CXL, trading latency for bandwidth and size.
Seems like a pretty good analogy as you admit there have been advancements while the basic structure has remained the same. Not sure why you have a pet peeve when it is highly analogous.
The ultimate shape of DRAM is the same, the main thing that's changed is the materials and techniques to produce it. Making it very impressive, but none the less completely recognizable by someone who was familiar with DRAM in the 70s.
The wheel is the same. Pluck someone from 1000 years ago and they'll be able to correctly identify a modern wheel even though they've never seen any of the composites that go into it. The function of the wheel is identical to how they used it.
You're right—the wheels on the Boeing 737 are, although made of forged aluminum or magnesium to withstand extreme force and heat, pretty much the same shape and operate in the same way as the Ljubljana Marshes Wheel of 3150 BCE.
Then again, flight itself has obviated—or, rather, introduced—many transit workloads that could be performed by wheeled vehicles, and operates on different principles entirely.
I found it fascinating that jetliner tires are designed to blow during extreme braking. An aborted takeoff (say microburst warning goes off), or an emergency landing will heat the brakes and tires to maximum. But there are plugs that will depressurize the tires without them coming apart. That way they don't catastrophically blow and spew tire pieces everywhere to get sucked into jet engines and damaging them.
pretty cool, sort of graceful (or at least planned) degradation of tires without cascading problems.
I have a feeling this will go the way of Optane and once it becomes useful they'll pull it and shelve it while keeping the patents/license of course so no one else can improve on it
cries I am sad at all of the cool designs that would work with optane if it had been widely adopted. We desperately need more points on the latency/cost per GB curve.
There doesn't seem to be much detail anywhere else either. All I was able to gather was that the memory dies are stacked (not new) but that the vias connecting the stack are angled instead of straight up and down and this is better because ... reasons?
I think the reasoning is better thermals and signal stability. The physics of the first seem to be that there is more metal to capture and distribute heat, but the signal integrity part beats me. By having better thermals, they can increase the memory clock and, thus, bandwidth and reduce latency.
«[T]he primary standout feature of this memory solution is the integration of a staggered interconnect topology that routes connections diagonally within the die stack rather than drilling straight down. According to Intel, the biggest benefit lies in ZAM's thermal capabilities.»
The connectors on the side indeed look like the letter Z. Maybe it disperses the stronger currents across the stack of the crystals, instead of concentrating.
I'd guess that it'd allow for thinner layers which is ultimately why you can pack in more memory.
And why it's not currently done is likely because it's hard enough to stack when everything is uniform. A small deformity in the first layer will spoil the entire chip.
This is not a drastically different technology like optane - it's almost solely a packaging change. It's applicable to exactly the same markets as normal DRAM, so if it dies customers will just switch to whatever DRAM variant wins instead.
Intel does these "throw spaghetti on the wall" kind of investments into potientially interesting companies/technologies all of the time - and have done so for decades.
Every time the recipient hypes the shit out of it, of course.
The main problem is that they often don't stick with it.
As far as I can tell, Intel more-or-less pioneered the idea of SSDs being the best storage rather than the cheap storage, for instance. The X25-M and X25-E were absurdly good. Then, once the market was established...they pulled out of it.
I’m still waiting for the Intel Arc B770 since a 5060 Ti and 9060 XT are already overpriced and if they just committed to something for once it wouldn’t be marginally worse.
Not that releasing the GPU would be something super innovative, they already have the B70.
Then, once the market was established...they pulled out of it.
This makes perfect sense given that Intel's target margins are pretty high. They only want to sell advanced tech, not commodities. Once SSDs became commoditized Intel was out.
Most of the big hit's in tech had a trendy index swinging moment, Intel has been searching for one for a long time since AMD64 undercut the Itanium. Hype drives a currently multi-billion dollar bubble. It's not always a bad idea to throw our holy noodles at the wall. You might find they hover is the sky and grow meatballs, could be big.
Well, peak weirdness was the thing involving Will-i-am from the Black Eyed Peas as a 'Futurist'/Spokesperson/IDEK.
I think what's semi-unfortunate is all the swings and misses, especially the cases where it wasn't necessarily a bad idea but Intel gives up too soon;
- Massively parallel simple-ish x86 cores a-la Xeon Phi; okay maybe not the best idea on the surface but I feel like nowadays the opportunities could be more forthcoming with how to reuse parts of that tech (And maybe they do but are just quiet about it... i.e. GPU acceleration)
- Optane. I think the tech would have been cheaper if they made terms for licensing easier, but maybe I'm missing part of the equation...
- This thing where they keep half assing the GPU strategy; Imagine if B70 launched last year alongside the B60 and B50, before DRAM prices went sideways. Or if they didn't take so long to release a >16GB GPU in the first place; that would have built a lot of interest, but instead they finally release a 32GB GPU alongside more bad news for the overall roadmap. The whole situation instead becomes a jarring rollercoaster that makes everyone worry that Intel is gonna kill the project the way everything but CPUs gets killed lately.
This wccf article also doesn't do a great job of describing, but the third slide it shows is very illustrative: rather than stack horizontally it stacks dram on its side. https://wccftech.com/intel-zam-memory-threatens-hbms-ai-thro...
I thought this was going to mean each stack was able to directly talk to the controller, since all stacks are resting on an interposer thing. But actually there is still a logic controller slice at the bottom of the stack, not at a right angle to the stack.
Instead of HBM microbumps between layers there is a more compact/dense TSV ("fusion bonded via-in-one") system. Intel once more showing their strong chiplet packing prowess! The claim is that thermals are still much better somehow, in spite of volumetric cell density increasing (from thinner layers). The demo has 8+1 dram+controller layers.
A search for patents by Stephen Morein who is listed as the CTO of SAIMEMORY shows this one, assigned to Intel, and published strangely close to the press release;
https://patents.google.com/patent/US20260040969A1/en
It talks there about 'Z axis memory' rather than angle, and that one is talking about inductive stuff through stacks of vias.
he's also got: https://patentsgazette.uspto.gov/week02/OG/html/1542-2/US125...
He obviously likes thinking about stacks of dies.
That slide is wrong, it's stacked the same way HBM is, the key difference is that they don't use microbumps anymore and instead have a hybrid bond system using copper to copper bonding. Copper is a good thermal conductor and the via is going through the entire wafer which in turn means that they are forming an uninterrupted copper cylinder that goes through all the layers. It's as if they put heat pipes inside the chip.
https://www.appliedmaterials.com/us/en/semiconductor/markets...
Edit: According to some rumors, the copper pillars go diagonally through the Z-Angle Memory.
The DRAM is not on its side though.
Apologies, yeah, that slide seems inaccurate now. They also discuss magnetically coupled wireless interface to the host, lol, yeah, no.
The newer slides show a more conventionally stacked die arrangement. With yet a third way of connecting stacks, that at least visually looks different than via-in-one columns and more like tabs between the layers, and definitely not wireless. Hard to guess what to expect from this assorted material.
Where is the music video for us who only want to learn about low level hardware through that medium?
Get perpendicular: https://www.dailymotion.com/video/x62mja
Off-topic, but 574 analytics and advertising partners is a few too many. Like… can you make money with just 200 ad partners?
Clearly not as much as you can with 574. Let's not be ridiculous.
I agree, insane.
That song is etched into my memory and bring back a flood of great feelings. Daily Motion is giving me a playback error, but here's a copy from youtube: https://www.youtube.com/watch?v=xb_PyKuI7II
I would love to be in the room when things like this are pitched. First, how high was the person that came up with the idea? Did the rest of the people get contact highs from how high this person clearly must have been to run with it? Also, what did the voice actors think? "Wait, this isn't for kids? Okay, it's your money"
It’s crazy that we have stalled on the structure of the basic DRAM cell for decades now.
Not that crazy. It's about the most basic structure you can make. Hard to make a better wheel.
The closest thing I can think of that's come close to maybe challenging DRAM is HP's memristors but those really didn't pan out (probably too much power consumption).
> Hard to make a better wheel.
Pet peeve: stupid analogy seeing how wheels kept being improved throughout the millennia with every new technology. The only thing in common is that it's round.
Similarly, DRAM in any way you see it has been improving to the point of barely being recognizable since the 70s.
That said, DIMMs and the whole bus idea is in dire need of getting a new type of bearing.
> That said, DIMMs and the whole bus idea is in dire need of getting a new type of bearing.
IBM has been using their own memory bus technology for both their POWER and Z machines. IIRC, it’s somewhat reminiscent of CXL, trading latency for bandwidth and size.
Seems like a pretty good analogy as you admit there have been advancements while the basic structure has remained the same. Not sure why you have a pet peeve when it is highly analogous.
The ultimate shape of DRAM is the same, the main thing that's changed is the materials and techniques to produce it. Making it very impressive, but none the less completely recognizable by someone who was familiar with DRAM in the 70s.
The wheel is the same. Pluck someone from 1000 years ago and they'll be able to correctly identify a modern wheel even though they've never seen any of the composites that go into it. The function of the wheel is identical to how they used it.
The latency of a DRAM cell has barely budged since the 1990s.
You're right—the wheels on the Boeing 737 are, although made of forged aluminum or magnesium to withstand extreme force and heat, pretty much the same shape and operate in the same way as the Ljubljana Marshes Wheel of 3150 BCE.
Then again, flight itself has obviated—or, rather, introduced—many transit workloads that could be performed by wheeled vehicles, and operates on different principles entirely.
I found it fascinating that jetliner tires are designed to blow during extreme braking. An aborted takeoff (say microburst warning goes off), or an emergency landing will heat the brakes and tires to maximum. But there are plugs that will depressurize the tires without them coming apart. That way they don't catastrophically blow and spew tire pieces everywhere to get sucked into jet engines and damaging them.
pretty cool, sort of graceful (or at least planned) degradation of tires without cascading problems.
I have a feeling this will go the way of Optane and once it becomes useful they'll pull it and shelve it while keeping the patents/license of course so no one else can improve on it
cries I am sad at all of the cool designs that would work with optane if it had been widely adopted. We desperately need more points on the latency/cost per GB curve.
The article says nothing about the construction or special qualities of ZAM, as compared to HBM :(
There doesn't seem to be much detail anywhere else either. All I was able to gather was that the memory dies are stacked (not new) but that the vias connecting the stack are angled instead of straight up and down and this is better because ... reasons?
I think the reasoning is better thermals and signal stability. The physics of the first seem to be that there is more metal to capture and distribute heat, but the signal integrity part beats me. By having better thermals, they can increase the memory clock and, thus, bandwidth and reduce latency.
Indeed, what is it? The article doesn't say, only espouses the supposed benefits.
Looks like an evolution of HBM.
https://wccftech.com/intel-showcases-its-zam-memory-prototyp...
«[T]he primary standout feature of this memory solution is the integration of a staggered interconnect topology that routes connections diagonally within the die stack rather than drilling straight down. According to Intel, the biggest benefit lies in ZAM's thermal capabilities.»
The connectors on the side indeed look like the letter Z. Maybe it disperses the stronger currents across the stack of the crystals, instead of concentrating.
I'd guess that it'd allow for thinner layers which is ultimately why you can pack in more memory.
And why it's not currently done is likely because it's hard enough to stack when everything is uniform. A small deformity in the first layer will spoil the entire chip.
But will this go the way of a “non core” product like Optane (or modems for that matter?)
This is not a drastically different technology like optane - it's almost solely a packaging change. It's applicable to exactly the same markets as normal DRAM, so if it dies customers will just switch to whatever DRAM variant wins instead.
Intel does these "throw spaghetti on the wall" kind of investments into potientially interesting companies/technologies all of the time - and have done so for decades.
Every time the recipient hypes the shit out of it, of course.
The main problem is that they often don't stick with it.
As far as I can tell, Intel more-or-less pioneered the idea of SSDs being the best storage rather than the cheap storage, for instance. The X25-M and X25-E were absurdly good. Then, once the market was established...they pulled out of it.
I’m still waiting for the Intel Arc B770 since a 5060 Ti and 9060 XT are already overpriced and if they just committed to something for once it wouldn’t be marginally worse.
Not that releasing the GPU would be something super innovative, they already have the B70.
Then, once the market was established...they pulled out of it.
This makes perfect sense given that Intel's target margins are pretty high. They only want to sell advanced tech, not commodities. Once SSDs became commoditized Intel was out.
It's possible that Intel wanted to seed the SSD industry.
They knew it wouldn't be profitable enough long term, but it would increase demand for their products.
An extreme and related example: https://en.wikipedia.org/wiki/Opticom_(company)
Popular science kind of backgrounder (can't vouch for the accuracy/relevancy - details are very scarce): https://www.geeksforgeeks.org/digital-logic/polymer-memory/
Most of the big hit's in tech had a trendy index swinging moment, Intel has been searching for one for a long time since AMD64 undercut the Itanium. Hype drives a currently multi-billion dollar bubble. It's not always a bad idea to throw our holy noodles at the wall. You might find they hover is the sky and grow meatballs, could be big.
Well, peak weirdness was the thing involving Will-i-am from the Black Eyed Peas as a 'Futurist'/Spokesperson/IDEK.
I think what's semi-unfortunate is all the swings and misses, especially the cases where it wasn't necessarily a bad idea but Intel gives up too soon;
- Massively parallel simple-ish x86 cores a-la Xeon Phi; okay maybe not the best idea on the surface but I feel like nowadays the opportunities could be more forthcoming with how to reuse parts of that tech (And maybe they do but are just quiet about it... i.e. GPU acceleration)
- Optane. I think the tech would have been cheaper if they made terms for licensing easier, but maybe I'm missing part of the equation...
- This thing where they keep half assing the GPU strategy; Imagine if B70 launched last year alongside the B60 and B50, before DRAM prices went sideways. Or if they didn't take so long to release a >16GB GPU in the first place; that would have built a lot of interest, but instead they finally release a 32GB GPU alongside more bad news for the overall roadmap. The whole situation instead becomes a jarring rollercoaster that makes everyone worry that Intel is gonna kill the project the way everything but CPUs gets killed lately.