From Newsgroup: comp.arch

This paper proposes using an optical fiber ring as a delay line to
store and distribute data in AI data centers. I think they're serious.

For you young folks, delay lines memories were common in the 1940s and
1950s before core memory replaced them, but they didn't use fiber.

Abstract

The rising pressure on DRAM availability and contract pricing reflects generative AI's massive high-performance memory requirements. This
pressure is heavily compounded by hyperscale data center expansion,
which now consumes a significant portion of global DRAM output. In
this work, we propose a new architecture: Fiber Memory, which
reimagines the role of optical fiber in a hyperscale data center,
deploying it as an active, recirculating delay-line memory for
immutable data, such as large language model (LLM) weights. We present
a data-parallel optical broadcast delay-line memory architecture that
accounts for fiber's physical realities. By incorporating
space-division multiplexed multi-core fibers (MCFs), passive optical tap-and-amplify interfaces, co-packaged optics (CPO), and regional
all-optical regeneration, our case study evaluation demonstrates that
Fiber Memory can eliminate redundant weight storage across 10,000 AI accelerators and reduce weight-delivery energy by over 70% compared to traditional HBM3e configurations.

https://arxiv.org/abs/2607.08407
--
Regards,
John Levine, johnl@taugh.com, Primary Perpetrator of "The Internet for Dummies",
Please consider the environment before reading this e-mail. https://jl.ly
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

John Levine <johnl@taugh.com> writes:

This paper proposes using an optical fiber ring as a delay line to
store and distribute data in AI data centers. I think they're serious.

For you young folks, delay lines memories were common in the 1940s and
1950s before core memory replaced them, but they didn't use fiber.

Abstract

The rising pressure on DRAM availability and contract pricing reflects >generative AI's massive high-performance memory requirements. This
pressure is heavily compounded by hyperscale data center expansion,
which now consumes a significant portion of global DRAM output. In
this work, we propose a new architecture: Fiber Memory, which
reimagines the role of optical fiber in a hyperscale data center,
deploying it as an active, recirculating delay-line memory for
immutable data, such as large language model (LLM) weights. We present
a data-parallel optical broadcast delay-line memory architecture that >accounts for fiber's physical realities. By incorporating
space-division multiplexed multi-core fibers (MCFs), passive optical >tap-and-amplify interfaces, co-packaged optics (CPO), and regional >all-optical regeneration, our case study evaluation demonstrates that
Fiber Memory can eliminate redundant weight storage across 10,000 AI >accelerators and reduce weight-delivery energy by over 70% compared to >traditional HBM3e configurations.

https://arxiv.org/abs/2607.08407

Interesting. From the summary, it's more like EPROM than DRAM.
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch


John Levine <johnl@taugh.com> posted:

This paper proposes using an optical fiber ring as a delay line to
store and distribute data in AI data centers. I think they're serious.

For you young folks, delay lines memories were common in the 1940s and
1950s before core memory replaced them, but they didn't use fiber.

Abstract

The rising pressure on DRAM availability and contract pricing reflects generative AI's massive high-performance memory requirements. This
pressure is heavily compounded by hyperscale data center expansion,
which now consumes a significant portion of global DRAM output.

Most of the memory being used for AI compute loads is necessary only
because of the interface between SSDs and DRAM. There are ways where
the same amount of data needs to be read off the SSDs, but less than
0.15% needs to arrive in DRAM.

For example, say one put FMAC units down on the SSD-FLASH controller
so that when SSD ships read-data, it ships it through said FMAC unit
to perform the cosine-similarity calculation and only ships only the
comparison result up to DRAM for CPU uses.

Vector length vary between 768 to 4096 elements (FP32) so cosine
results are between 0.15% and 0.025% of vector data size.

In
this work, we propose a new architecture: Fiber Memory, which
reimagines the role of optical fiber in a hyperscale data center,
deploying it as an active, recirculating delay-line memory for
immutable data, such as large language model (LLM) weights. We present
a data-parallel optical broadcast delay-line memory architecture that accounts for fiber's physical realities. By incorporating
space-division multiplexed multi-core fibers (MCFs), passive optical tap-and-amplify interfaces, co-packaged optics (CPO), and regional all-optical regeneration, our case study evaluation demonstrates that
Fiber Memory can eliminate redundant weight storage across 10,000 AI accelerators and reduce weight-delivery energy by over 70% compared to traditional HBM3e configurations.

https://arxiv.org/abs/2607.08407

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

This paper proposes using an optical fiber ring as a delay line to
store and distribute data in AI data centers. I think they're serious.
For you young folks, delay lines memories were common in the 1940s and
1950s before core memory replaced them, but they didn't use fiber.

Right, but this is specially designed for the case where "every"
processor in your large system needs repeatedly the same sequence
of data. That seems potentially applicable to today's LLMs, but it's
*highly* specialized.

I don't like the direction this is going since it seems to imply that
running such LLMs would be inherently limited to large data-centers,
thereby taking control out of the end-users hands.


=== Stefan
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

In article <jwvpl0umsyt.fsf-monnier+comp.arch@gnu.org>, monnier@iro.umontreal.ca (Stefan Monnier) wrote:

I don't like the direction this is going since it seems to imply
that running such LLMs would be inherently limited to large
data-centers, thereby taking control out of the end-users hands.

That seems entirely consistent with the agendas of the companies running
most of the world's LLMs.

John
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

According to Stefan Monnier <monnier@iro.umontreal.ca>:

This paper proposes using an optical fiber ring as a delay line to
store and distribute data in AI data centers. I think they're serious.
For you young folks, delay lines memories were common in the 1940s and
1950s before core memory replaced them, but they didn't use fiber.

Right, but this is specially designed for the case where "every"
processor in your large system needs repeatedly the same sequence
of data. That seems potentially applicable to today's LLMs, but it's >*highly* specialized.

I don't like the direction this is going since it seems to imply that
running such LLMs would be inherently limited to large data-centers,
thereby taking control out of the end-users hands.

I don't think it's much of a change from the current situation, since
the fiber loop is intended to replace gigabyte HBMs which already only
can work in large data centers. If they needed fewer HBMs so the RAM manufacturers could make more DRAM and SD, we'd all be better off.

This is separate from the issue of whether you really need a giant
data centre to do useful inference, or the Apple model where you do
it mostly on your device and only call out to the cloud occasionally
will work.
--
Regards,
John Levine, johnl@taugh.com, Primary Perpetrator of "The Internet for Dummies",
Please consider the environment before reading this e-mail. https://jl.ly
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

On Fri, 10 Jul 2026 02:42:59 -0000 (UTC), John Levine
<johnl@taugh.com> wrote:

This paper proposes using an optical fiber ring as a delay line to
store and distribute data in AI data centers. I think they're serious.

For you young folks, delay lines memories were common in the 1940s and
1950s before core memory replaced them, but they didn't use fiber.

In the 70's and 80's there was considerable interest in magnetic
bubble memory which essentially is a delay line technology. It was
expensive, and bit density was a problem, but since it was
non-volatile, it often was used in machines designed for harsh
environments.

In the 80's Olivetti sold as a (IBM)PC expansion board a solid state
drive that used bubble memory.

In the 2K's IBM's "racetrack" memory was a form of bubble memory.
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

On Sat, 11 Jul 2026 14:44 +0100 (BST), John Dallman wrote:

That seems entirely consistent with the agendas of the companies
running most of the world's LLMs.

On the bright side, they seem to be having real trouble making money
from the massive investments and expenditure being made in equipment
and software/training and, in particular, on electricity bills.

The pesky users donrCOt seem to want to pay $-ure4-N anywhere near what it costs to provide the service.
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

John Levine [2026-07-10 02:42:59] wrote:

This paper proposes using an optical fiber ring as a delay line to
store and distribute data in AI data centers. I think they're serious.

BTW, I think the core of the idea is to recognize that some of the
workload in large AI datacenters is not just "SIMT" but that some part
of the data is exactly the same for some of those threads (e.g. say most
of the threads are performing a "vector x matrix" multiply with
a different vector each but all with the same matrix), so you might be
able to get similar benefits with some kind of "broadcast"
DRAM/SSD reads.


=== Stefan
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch


Stefan Monnier <monnier@iro.umontreal.ca> posted:

John Levine [2026-07-10 02:42:59] wrote:

This paper proposes using an optical fiber ring as a delay line to
store and distribute data in AI data centers. I think they're serious.

BTW, I think the core of the idea is to recognize that some of the
workload in large AI datacenters is not just "SIMT" but that some part
of the data is exactly the same for some of those threads (e.g. say most
of the threads are performing a "vector x matrix" multiply with
a different vector each but all with the same matrix), so you might be
able to get similar benefits with some kind of "broadcast"
DRAM/SSD reads.

Much of the data is in an HNSW tree, each unit consisting of a vector
between 768-and-8192 FP32 elements. Cosine-similarity is simply a
weighted inner-product. Each node in the tree has an in-plane fan-out
and a next-level fan-out (pointers). Any inner-product with greater
than 80% cs-score is kept and used to access the next layer down.

With 100,000,000 vectors in the HNSW tree one needs to perform about
2,200 cosine-similarities and about 7 layers in the tree (~30TB), 1,000,000,000 needs 5,000 c-s and 8 layers in the tree (~300TB).
Ideally, the whole VDB resides in DRAM--but alas...
Longer vectors need fewer layers in the tree.

So, every query to the vector-data-base needs a cosine-similarity to
all of the elements at the top level. and those with promising c-ss
then access the next layer down in the tree. By restricting the lower
layers of the tree to only promising queries is how the HNSW tree
reduces the number of c-s it needs to perform.

The construction of the tree is interesting in its own light, especially
when other queries are accessing while a new data-point is being installed.

=== Stefan

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

George Neuner <gneuner2@comcast.net> wrote:

On Fri, 10 Jul 2026 02:42:59 -0000 (UTC), John Levine
<johnl@taugh.com> wrote:

[ ... ]
For you young folks, delay lines memories were common in the 1940s and >1950s before core memory replaced them, but they didn't use fiber.

In the 70's and 80's there was considerable interest in magnetic
bubble memory which essentially is a delay line technology. It was expensive, and bit density was a problem, but since it was
non-volatile, it often was used in machines designed for harsh
environments.

Nitpick: bubble memories were built on shift registers, but they
were not delay lines. If you stopped the rotating magnetic field,
the bits stopped circulating. Delay line memories have no equivalent.
--
pa at panix dot com
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

On Sun, 12 Jul 2026 22:42:00 -0000 (UTC), Pierre Asselin wrote:

Nitpick: bubble memories were built on shift registers, but they
were not delay lines. If you stopped the rotating magnetic field,
the bits stopped circulating. Delay line memories have no
equivalent.

Kinda like the difference between DRAM and SRAM, I guess ...
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

On 7/11/2026 5:19 PM, George Neuner wrote:

On Fri, 10 Jul 2026 02:42:59 -0000 (UTC), John Levine
<johnl@taugh.com> wrote:

This paper proposes using an optical fiber ring as a delay line to
store and distribute data in AI data centers. I think they're serious.

For you young folks, delay lines memories were common in the 1940s and
1950s before core memory replaced them, but they didn't use fiber.

In the 70's and 80's there was considerable interest in magnetic
bubble memory which essentially is a delay line technology. It was expensive, and bit density was a problem, but since it was
non-volatile, it often was used in machines designed for harsh
environments.

Yes.

And in the mid to late 1970s, a company I later worked for sold an SSD
(Solid State *Drum* - a PCM replacement for a *real* drum peripheral
memory. Not a head per track disk that a certain vendor called a "drum"
- sorry my pet peeve.) It originally used CCD chips, similar to bubble
memory but faster, though totally different technology. However, CCD
memory was volatile, which they overcame by having a UPS. Soon the
economies of DRAM overtook the CCD chips and replaced them. The UPS was
still there to compensate for the volatility. IIRC the replacement
DRAMS were 4K chips.
--
- Stephen Fuld
(e-mail address disguised to prevent spam)
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

On 2026-07-13 1:42, Pierre Asselin wrote:

George Neuner <gneuner2@comcast.net> wrote:

On Fri, 10 Jul 2026 02:42:59 -0000 (UTC), John Levine
<johnl@taugh.com> wrote:

[ ... ]
For you young folks, delay lines memories were common in the 1940s and
1950s before core memory replaced them, but they didn't use fiber.

In the 70's and 80's there was considerable interest in magnetic
bubble memory which essentially is a delay line technology. It was
expensive, and bit density was a problem, but since it was
non-volatile, it often was used in machines designed for harsh
environments.

Nitpick: bubble memories were built on shift registers, but they
were not delay lines. If you stopped the rotating magnetic field,
the bits stopped circulating. Delay line memories have no equivalent.

One could make an argument that all continuously rotating magnetic media (drums, disks) are also "delay lines"; does it really make a difference
if the material carrying the binary pulses also moves?

Rotating magnetic media have the same basic property as delay lines:
that a particular storage location can be accessed only when it is
passing by the read/write apparatus, and the time between such moments
is constant, and can be inconveniently long for the programmer's purposes.

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

Stephen Fuld <sfuld@alumni.cmu.edu.invalid> wrote:

On 7/11/2026 5:19 PM, George Neuner wrote:

On Fri, 10 Jul 2026 02:42:59 -0000 (UTC), John Levine
<johnl@taugh.com> wrote:

This paper proposes using an optical fiber ring as a delay line to
store and distribute data in AI data centers. I think they're serious.

For you young folks, delay lines memories were common in the 1940s and
1950s before core memory replaced them, but they didn't use fiber.

In the 70's and 80's there was considerable interest in magnetic
bubble memory which essentially is a delay line technology. It was
expensive, and bit density was a problem, but since it was
non-volatile, it often was used in machines designed for harsh
environments.

Yes.

And in the mid to late 1970s, a company I later worked for sold an SSD (Solid State *Drum* - a PCM replacement for a *real* drum peripheral
memory. Not a head per track disk that a certain vendor called a "drum"
- sorry my pet peeve.) It originally used CCD chips, similar to bubble memory but faster, though totally different technology. However, CCD
memory was volatile, which they overcame by having a UPS. Soon the economies of DRAM overtook the CCD chips and replaced them. The UPS was still there to compensate for the volatility. IIRC the replacement
DRAMS were 4K chips.

Not 64k bits? IIUC in earlier period CCD had density advantage, but
that vanished when 64k bits DRAM became available.
--
Waldek Hebisch
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

On 7/13/2026 1:13 PM, Waldek Hebisch wrote:

Stephen Fuld <sfuld@alumni.cmu.edu.invalid> wrote:

On 7/11/2026 5:19 PM, George Neuner wrote:

On Fri, 10 Jul 2026 02:42:59 -0000 (UTC), John Levine
<johnl@taugh.com> wrote:

This paper proposes using an optical fiber ring as a delay line to
store and distribute data in AI data centers. I think they're serious. >>>>
For you young folks, delay lines memories were common in the 1940s and >>>> 1950s before core memory replaced them, but they didn't use fiber.

In the 70's and 80's there was considerable interest in magnetic
bubble memory which essentially is a delay line technology. It was
expensive, and bit density was a problem, but since it was
non-volatile, it often was used in machines designed for harsh
environments.

Yes.

And in the mid to late 1970s, a company I later worked for sold an SSD
(Solid State *Drum* - a PCM replacement for a *real* drum peripheral
memory. Not a head per track disk that a certain vendor called a "drum"
- sorry my pet peeve.) It originally used CCD chips, similar to bubble
memory but faster, though totally different technology. However, CCD
memory was volatile, which they overcame by having a UPS. Soon the
economies of DRAM overtook the CCD chips and replaced them. The UPS was
still there to compensate for the volatility. IIRC the replacement
DRAMS were 4K chips.

Not 64k bits? IIUC in earlier period CCD had density advantage, but
that vanished when 64k bits DRAM became available.

You may very well be right. As I said, it was before I joined the
company, and it was nearly 50 years ago. :-(
--
- Stephen Fuld
(e-mail address disguised to prevent spam)
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch


Stephen Fuld <sfuld@alumni.cmu.edu.invalid> posted:

On 7/13/2026 1:13 PM, Waldek Hebisch wrote:

Stephen Fuld <sfuld@alumni.cmu.edu.invalid> wrote:

On 7/11/2026 5:19 PM, George Neuner wrote:

On Fri, 10 Jul 2026 02:42:59 -0000 (UTC), John Levine
<johnl@taugh.com> wrote:

This paper proposes using an optical fiber ring as a delay line to
store and distribute data in AI data centers. I think they're serious. >>>>
For you young folks, delay lines memories were common in the 1940s and >>>> 1950s before core memory replaced them, but they didn't use fiber.

In the 70's and 80's there was considerable interest in magnetic
bubble memory which essentially is a delay line technology. It was
expensive, and bit density was a problem, but since it was
non-volatile, it often was used in machines designed for harsh
environments.

Yes.

And in the mid to late 1970s, a company I later worked for sold an SSD
(Solid State *Drum* - a PCM replacement for a *real* drum peripheral
memory. Not a head per track disk that a certain vendor called a "drum" >> - sorry my pet peeve.) It originally used CCD chips, similar to bubble
memory but faster, though totally different technology. However, CCD
memory was volatile, which they overcame by having a UPS. Soon the
economies of DRAM overtook the CCD chips and replaced them. The UPS was >> still there to compensate for the volatility. IIRC the replacement
DRAMS were 4K chips.

Not 64k bits? IIUC in earlier period CCD had density advantage, but
that vanished when 64k bits DRAM became available.

CCDs have an advantage no DRAM ever could:: the charge was analog !

A friend of mine had a stereo delay line which we used to alter the
"size" of the room with the speakers at the back being delayed by
20-50 ms (depending on the kind of music.) The effect was interesting
but only partially successful.

You may very well be right. As I said, it was before I joined the
company, and it was nearly 50 years ago. :-(

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

Niklas Holsti <niklas.holsti@tidorum.invalid> wrote:

Rotating magnetic media have the same basic property as delay lines:
that a particular storage location can be accessed only when it is
passing by the read/write apparatus, and the time between such moments
is constant, and can be inconveniently long for the programmer's purposes.

Rotating magnetic media (and magnetic bubbles) are nonvolatile storage.
Delay lines are volatile: if you don't detect, re-shape and
re-broadcast the signal after its round trip, the information is lost.
--
pa at panix dot com
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

On 2026-Jul-11 20:19, George Neuner wrote:

In the 70's and 80's there was considerable interest in magnetic
bubble memory which essentially is a delay line technology. It was expensive, and bit density was a problem, but since it was
non-volatile, it often was used in machines designed for harsh
environments.

In the 80's Olivetti sold as a (IBM)PC expansion board a solid state
drive that used bubble memory.

In the 2K's IBM's "racetrack" memory was a form of bubble memory.

The Intel 1Mb bubble memories were build using
X-ray lithography from a synchrotron.
Might account for the cost.



--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

On Tue, 14 Jul 2026 00:39:25 GMT, MitchAlsup wrote:

A friend of mine had a stereo delay line which we used to alter the
"size" of the room with the speakers at the back being delayed by
20-50 ms (depending on the kind of music.) The effect was
interesting but only partially successful.

Was the room acoustically rCLdeadrCY? (Soft padding, curtains etc.)
Otherwise its own reverb would likely be confusing the issue.
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

On 2026-07-14 3:56, Pierre Asselin wrote:

Niklas Holsti <niklas.holsti@tidorum.invalid> wrote:

Rotating magnetic media have the same basic property as delay lines:
that a particular storage location can be accessed only when it is
passing by the read/write apparatus, and the time between such moments
is constant, and can be inconveniently long for the programmer's purposes.

Rotating magnetic media (and magnetic bubbles) are nonvolatile storage.
Delay lines are volatile: if you don't detect, re-shape and
re-broadcast the signal after its round trip, the information is lost.

Certainly, but that is a detail of the HW implementation. Dynamic RAM is
still Random Access memory, as was magnetic-core memory which had to be rewritten after every read, because reads were destructive.


--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

On 2026-07-14 12:46, Niklas Holsti wrote:

On 2026-07-14 3:56, Pierre Asselin wrote:

Niklas Holsti <niklas.holsti@tidorum.invalid> wrote:

Rotating magnetic media have the same basic property as delay lines:
that a particular storage location can be accessed only when it is
passing by the read/write apparatus, and the time between such moments
is constant, and can be inconveniently long for the programmer's
purposes.

Rotating magnetic media (and magnetic bubbles) are nonvolatile storage.
Delay lines are volatile: if you don't detect, re-shape and
re-broadcast the signal after its round trip, the information is lost.

Certainly, but that is a detail of the HW implementation. Dynamic RAM is still Random Access memory, as was magnetic-core memory which had to be rewritten after every read, because reads were destructive.

I should add that of course the nonvolatility of rotating magnetic media
is (or perhaps better, was) very important for their uses. Their "delay
line" nature (serial access) was mitigated and became less of a problem
when RAM became available in large amounts for track buffering and memory-resident databases, leaving the nonvolatility as the main purpose
of magnetic media, whether rotating or random access (MRAM).

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch


Lawrence =?iso-8859-13?q?D=FFOliveiro?= <ldo@nz.invalid> posted:

On Tue, 14 Jul 2026 00:39:25 GMT, MitchAlsup wrote:

A friend of mine had a stereo delay line which we used to alter the
"size" of the room with the speakers at the back being delayed by
20-50 ms (depending on the kind of music.) The effect was
interesting but only partially successful.

Was the room acoustically rCLdeadrCY? (Soft padding, curtains etc.)
Otherwise its own reverb would likely be confusing the issue.

Typical off-campus college apartment for 4 people. So, no.
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

Niklas Holsti <niklas.holsti@tidorum.invalid> wrote:

On 2026-07-14 3:56, Pierre Asselin wrote:

Niklas Holsti <niklas.holsti@tidorum.invalid> wrote:

Rotating magnetic media have the same basic property as delay lines:
that a particular storage location can be accessed only when it is
passing by the read/write apparatus, and the time between such moments
is constant, and can be inconveniently long for the programmer's purposes.

Rotating magnetic media (and magnetic bubbles) are nonvolatile storage. Delay lines are volatile: if you don't detect, re-shape and
re-broadcast the signal after its round trip, the information is lost.

Certainly, but that is a detail of the HW implementation. Dynamic RAM is still Random Access memory, as was magnetic-core memory which had to be rewritten after every read, because reads were destructive.

It's still a far-fetched analogy (between delay lines and, e.g.
disks). With a disk, you can park the heads and stop the rotation,
the data stays in place. With bubbles, you can stop the rotating field
and the data stays in place. There is no stopping a delay line.
--
pa at panix dot com
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

EricP <ThatWouldBeTelling@thevillage.com> wrote:

The Intel 1Mb bubble memories were build using
X-ray lithography from a synchrotron.
Might account for the cost.

I find that hard to believe. The minimum feature wasn't that small.
--
pa at panix dot com
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

On 2026-Jul-20 15:02, Pierre Asselin wrote:

EricP <ThatWouldBeTelling@thevillage.com> wrote:

The Intel 1Mb bubble memories were build using
X-ray lithography from a synchrotron.
Might account for the cost.

I find that hard to believe. The minimum feature wasn't that small.

Searching on " "bubble memory" synchrotron"
gives a number of hits circa 1975 but most are paywalled.

https://scholar.google.com/scholar?hl=en&as_sdt=0%2C5&q=%22bubble+memory%22+synchrotron&btnG=

X-ray lithography for VLSI, 1983 https://ieeexplore.ieee.org/abstract/document/1456910/
the google summary says:
"Most of our X-ray experience is with the Intel 1-Mbit bubble memory (Synchrotron radiation sources emit almost parallel X-ray"

This IBM paper is available:

Early history of X-ray lithography at IBM, 1993, E. Spiller https://mirrors.meulie.net/bitsavers.org/pdf/ibm/IBM_Journal_of_Research_and_Development/373/ibmrd3703C.pdf

"Bubble memoy and other applications
Starting in 1974, we delivered resist patterns to the bubble memory project.
We were all very excited, because we believed that bubble memories could
become an important product, and we were sure that we could supply the lithographic tool to manufacture these devices, to linewidths below
0.1 um. However, the enthusiasm for bubble memories faded, and plans
to develop a product were canceled in 1976."

As to feature size, circa 1975-80 I think they were still using contact/proximity lithography. Projection lithography came
with the introduction of the wafer stepper in 1978,
though it says they really only got going in the 1980's.
"Proximity printing had poorer resolution than contact printing
(due to the gap allowing more diffraction to occur) but generated
far fewer defects. The resolution was sufficient for down to 2 micrometre production. In 1978, the step-and-repeat projection system appeared." https://en.wikipedia.org/wiki/Contact_lithography https://en.wikipedia.org/wiki/Stepper

--- Synchronet 3.22a-Linux NewsLink 1.2