Send Link mailing list submissions to
[email protected]
To subscribe or unsubscribe via the World Wide Web, visit
https://mailman.anu.edu.au/mailman/listinfo/link
or, via email, send a message with subject or body 'help' to
[email protected]
You can reach the person managing the list at
[email protected]
When replying, please edit your Subject line so it is more specific
than "Re: Contents of Link digest..."
Today's Topics:
1. The China Chip Chokes (Stephen Loosley)
2. Re: The China Chip Chokes (Antony Barry)
3. The Open Source AI Definition ? 1.0 (Stephen Loosley)
----------------------------------------------------------------------
Message: 1
Date: Tue, 29 Oct 2024 11:56:01 +1030
From: Stephen Loosley <[email protected]>
To: "link" <[email protected]>
Subject: [LINK] The China Chip Chokes
Message-ID: <[email protected]>
Content-Type: text/plain; charset="UTF-8"
The China Chip Chokepoints
The U.S. and its allies have leverage in the competition to build advanced
microchips.
By Matt Brazil, Matt Bruzzese and Peter W. Singer October 27, 2024
https://www.defenseone.com/ideas/2024/10/chinas-chip-chokepoints/400593/
Earlier this year, American investigators were tipped off about a plan to ship
semiconductor manufacturing equipment to China.
The alarm was sounded after two Chicago-based buyers of a Dynatex DTX-MDB 150
scribe and break machine asked the California manufacturer to file export
paperwork that omitted the final destination: Chengdu GaStone Technology
Company.
The firm in southwest China is under U.S. sanction for an illegal 2021 scheme
to purchase military-grade integrated circuits.
In April, Lin Chen was arrested and charged while Han Li escaped to China.
The case illuminates the complex U.S.-China competition to produce
ever-more-advanced chips. CGTC sought the Dynatex machine to handle a final
step in making military-grade gallium arsenide chips, which can better
withstand extreme temperatures and radiation than ordinary silicon devices.
Had they succeeded, the Chengdu firm could have improved their
chip-manufacturing "success rate" and therefore their ability to meet customer
demand?including from the Chinese military.
While companies such as SMIC have turned China into a chip-manufacturing
powerhouse, the country remains unable to mass-produce the most advanced chips
on its own.
Despite massive government investment, China continues to rely on foreign
companies for the chips that power everything from the AI revolution to the
internet of things to the most advanced military equipment.
This problem remains a huge vulnerability for China?s economy, its military,
and its global ambitions.
Commentators inside China identify advanced chips as one of the chokepoint
technologies that keep the country from technological parity with the most
advanced world economies. In 2018, the government S&T Daily publication named
35 such technologies, most in the fields of information technology,
electronics, manufacturing, biotech, and new-energy vehicles.
Eight of these chokepoints related to the design and manufacture of advanced
chips, casting doubt on the Chinese ability to produce such chips anytime soon.
Western leverage
While this China struggle is hardly secret, few stop to think about what
exactly it means, why it is so difficult to overcome?and what leverage it might
give U.S. and allied governments.
To paraphrase Boromir, one does not simply insert silicon into one end of a
machine and wait for an advanced chip to emerge from the other. Producing these
chips is a complex, multi-stage process. And each stage, in turn, provides
numerous avenues for U.S. and allied governments to apply pressure on Chinese
efforts at its own advanced chip production.
Before the chips can even be manufactured, they must be designed. This
necessitates the use of electronic design automation software, which brings
order to the billions of transistors that go into a modern chip, a task
impossible for human engineers to accomplish on their own.
Here lies the Chinese first chokepoint.
Chinese-made EDA software from companies such as Beijing Empyrean and Primarius
Technologies can handle analog designs and simpler IoT applications but appear
unable to compete at the high end. Thus, the Chinese market for this
specialized software is dominated by the three largest U.S. developers:
Synopsys, Cadence, and Siemens EDA, who collectively hold 90 percent of the PRC
market.
In 2022, the U.S. Commerce Department placed export controls on the most
advanced versions of this software, which is needed to design top-of-the-line
3nm chips.
Next, raw materials must be turned into the silicon wafers that will eventually
become chips. Silicon is extracted from sand, refined to 99.9999999% purity,
melted, and fashioned into a single-crystal ingot. This ingot, called a
?boule,? is sliced into thin wafers and polished to create an ultra-flat,
mirror-like surface.
To achieve the extraordinary nanometer-level smoothness that is required,
ultra-precise polishing equipment is needed. Herein lies a second chokepoint:
China appears to lack companies that can produce this equipment, although their
research is showing some promising results. The leading manufacturers are in
the U.S. and Germany.
In 2020, the U.S. government slapped export controls on the most advanced
versions of this equipment.
Once the wafers are ready, one side is coated with a light-sensitive chemical
called photoresist to prepare it for photolithography.
Production of this chemical is a third chokepoint. Extreme-ultraviolet
photoresist is needed for the densest chips below 20nm, and this market is
dominated by Japanese companies bound by strict export-control policies.
Even South Korean and Taiwanese semiconductor manufacturers depend on Japanese
advanced resists. Chinese companies make less-advanced photoresists, which are
in high demand worldwide for manufacturing inexpensive commodity chips such as
those found in automobiles and consumer products. But their products are
insufficient for leading-edge devices.
Chinese firms such as Jingrui, Red Avenue, and its subsidiary Beijing Kempur, a
U.S.-China joint venture, are pouring research into making advanced
photoresists.
This will not be easy, because older photoresist chemicals based on bis-azide
photoresists cannot create circuits below two micrometers (2,000 nanometers).
Further, the most advanced core technologies, including fine-particle filtering
for new chemicals such as fullerenes (a zero-dimension carbon nanomaterial for
high-resolution photoresists) may not be exported from the U.S. and its allies
to China without a license.
Next, the light-sensitive wafers are sent through a photolithography machine,
which projects light through a series of reticles to produce microscopic
circuit patterns.
Herein lies yet another chokepoint. The most advanced machines, manufactured by
Dutch company ASML, have more than 100,000 parts and cost hundreds of millions
of dollars. They have carried a presumption-of-denial in the export application
process led by Washington since 2022.
While China manufactures huge amounts of less advanced chips, it remains
incapable of reproducing the complex photolithography machinery needed for the
most advanced devices.
Based on past performance, the prospects for quickly catching up with the
world?s photolithography industry leaders are not good: China has striven to do
so since the 1980s and has historically been a decade or two behind.
The leading Chinese producer, SMIC, can reportedly make 28nm chips, far from
the global leading edge of 2 to 3nm.
And while China is reportedly now producing a 7nm chip for the latest Huawei
smartphone, this is believed to use a less advanced and less efficient form of
lithography from a machine acquired from ASML before U.S. controls went into
effect.
The SMIC success rate for chips on each wafer may be low and commercially
unacceptable, keeping costs high and production limited.
And Washington is trying to further restrict sales and service of the
lithography machines SMIC used to make them.
There are almost certainly other chokepoints buried in the intricate process
that produces modern semiconductors, perhaps in the stages of etching, ion
implantation, metal deposition, dicing, and final assembly and testing.
For instance, the Chen and Li efforts in Chicago indicate that China depends on
foreign scribe-and-break machines to dice military-grade gallium arsenide
chips, making this yet another potential chokepoint not named in the S&T Daily
2018 list.
The multiple chokepoints China must contend with provide a dual picture. They
show the difficulties it faces in its quest for world-class chips.
Meanwhile, they show how Washingtons efforts to keep cutting-edge semiconductor
technology out of China have achieved some successes, but still face
significant hurdles. Export controls have only been partly updated since the
Cold War and rely on a network of nations and companies. The United States
needs a new multilateral export regime powered by consensus among key partners
that make semiconductor manufacturing tools, many of whom would prefer to
continue selling to China at some level.
China will continue to improve its chip-making through overt and covert means,
for the one certainty about the future of the economy and warfare is that they
will be powered by semiconductors.
The question is where those technologies will come from, and how much volume in
chips and equipment clandestine actors can manage to smuggle into China.
--
Matt Brazil is a China Analyst with BluePath Labs and the co-author (with Peter
Mattis) of Chinese Communist Espionage, An Intelligence Primer (2019). He is a
former employee of the Commerce Department and of Intel Corporate Security. His
views do not represent those of any institution.
Matt Bruzzese is a senior Chinese-language analyst for BluePath Labs.
P.W. Singer is a best-selling author of such books on war and technology as
Wired for War, Ghost Fleet, and Burn-In; senior fellow at New America; and
co-founder of Useful Fiction, a strategic narratives company
------------------------------
Message: 2
Date: Tue, 29 Oct 2024 14:58:51 +1100
From: Antony Barry <[email protected]>
To: Stephen Loosley <[email protected]>
Cc: link <[email protected]>
Subject: Re: [LINK] The China Chip Chokes
Message-ID:
<CAECOtWz2oJMwfGqJsPyLqKC7y3mBiLb=qluvejcxwtuvuev...@mail.gmail.com>
Content-Type: text/plain; charset="UTF-8"
On Tue, Oct 29, 2024 at 12:30?PM Stephen Loosley <[email protected]>
wrote:
>
> To paraphrase Boromir, one does not simply insert silicon into one end of
> a machine and wait for an advanced chip to emerge from the other.
>
>
While I have read the Lord of the Rings (twice) and seen the films I could
not remember the quotation. So I asked the AI Perplexity and got the
following fulsome answer -
The quotation you're referring to is paraphrasing Boromir's famous line
> from The Lord of the Rings: The Fellowship of the Ring: "One does not
> simply walk into Mordor"[1][2]. In the original context, Boromir is
> describing the extreme difficulty and danger of entering Mordor, the realm
> of the dark lord Sauron.
>
> The paraphrase you provided adapts this quote to discuss the complexity of
> semiconductor manufacturing. Just as Boromir emphasizes that entering
> Mordor is not a simple task, the paraphrase highlights that producing
> advanced chips is not a straightforward process. It cleverly draws a
> parallel between the challenges of Frodo's quest and the intricacies of
> modern technology production.
>
> The original quote continues to describe the dangers and obstacles in
> Mordor, much like how the paraphrase goes on to explain the multi-stage
> nature of chip production and the potential for intervention by
> governments. This adaptation effectively uses the well-known quote to
> illustrate the complexity and geopolitical implications of advanced
> semiconductor manufacturing.
>
> Sources
> [1] Sean Bean as Boromir - The Fellowship of the Ring (2001) - IMDb
> https://www.imdb.com/title/tt0120737/characters/nm0000293
> [2] The Lord of the Rings: The Fellowship of the Ring (2001) - Quotes
> https://www.imdb.com/title/tt0120737/quotes/?item=qt0446011
> [3] One Does Not Simply Walk Into Mordor - Know Your Meme
> https://knowyourmeme.com/memes/one-does-not-simply-walk-into-mordor
> [4] The Story Behind Your Favorite 'Lord of The Rings' Meme - Newsweek
> https://www.newsweek.com/lord-rings-meme-boromir-one-does-not-simply-walk-mordor-fellowship-sean-bean-1507844
>
--
Mob:04 3365 2400 Email: [email protected], [email protected]
------------------------------
Message: 3
Date: Tue, 29 Oct 2024 15:32:39 +1030
From: Stephen Loosley <[email protected]>
To: "link" <[email protected]>
Subject: [LINK] The Open Source AI Definition ? 1.0
Message-ID: <[email protected]>
Content-Type: text/plain; charset="UTF-8"
What Does Open-Source AI Actually Mean .. There is Finally a Definition
Companies like Meta have been calling their products open source. They're not.
By Todd Feathers Published October 28, 2024
https://gizmodo.com/what-does-open-source-ai-actually-mean-theres-finally-a-definition-2000517385
[Photo caption: Meta has claimed its Llama models are open source, but they
don't meet the Open Source Initiative's definition.
? Anadolu/Getty Images]
In the buzzy world of AI, boring things like definitions often get overlooked.
The term artificial intelligence itself is so broadly applied that it can refer
to everything from linear regression models to killer robots.
But when it comes to regulating emerging technologies, clear and precise
definitions are important. Without them, you end up with the kind of
goofy-if-it-wasnt-so-serious debates that state lawmakers around the country
are having, like whether the language they wrote to ban deceptive deepfakes
will also apply to spell check.
So it is notable that, following years of research and global debate, the Open
Source Initiative has finally agreed on a definition for Open Source AI that
the nonprofit organization hopes can guide international regulation.
In order to be labeled open source under the new definition, an AI
system?including its component code, weights, and training data?must be made
freely available in such a way that anyone can, without permission, use it for
any purpose, study how it works, modify it, and share it with others.
That is a pretty big departure from the way some tech companies have used the
label amid the generative AI arms race.
Most notably, Meta advertises its Llama family of models as open-source because
they?re free to use (as long as developers adhere to the company?s license
terms) and some of the code is publicly available.
Last year, Metal also helped create a lobbying coalition called the AI Alliance
to advocate for policies that benefit its particular brand of open-source
technologies.
https://apnews.com/article/ai-opensource-meta-ibm-chatgpt-dd61e99ac8135b36872b3987601067ec
Llama models do not qualify as open source because their licenses still place
limits on how they can be used for some commercial purposes, like improving
other large language models. They also outright prohibit uses that might
violate various laws or cause harm. Meta has also not fully disclosed the
training data for its Llama models.
----
The Open Source Initiative
https://opensource.org/ai/open-source-ai-definition
The Open Source AI Definition ? 1.0
Endorse the Open Source AI Definition: have your organization appended to the
list of supporters of version 1.0
https://opensource.org/ai/endorsements
Preamble
Why we need Open Source Artificial Intelligence (AI)
Open Source has demonstrated that massive benefits accrue to everyone after
removing the barriers to learning, using, sharing and improving software
systems.
These benefits are the result of using licenses that adhere to the Open Source
Definition. For AI, society needs at least the same essential freedoms of Open
Source to enable AI developers, deployers and end users to enjoy those same
benefits: autonomy, transparency, frictionless reuse and collaborative
improvement.
What is Open Source AI
When we refer to a ?system,? we are speaking both broadly about a fully
functional structure and its discrete structural elements. To be considered
Open Source, the requirements are the same, whether applied to a system, a
model, weights and parameters, or other structural elements.
An Open Source AI is an AI system made available under terms and in a way that
grant the freedoms to:
* Use the system for any purpose and without having to ask for permission.
* Study how the system works and inspect its components.
* Modify the system for any purpose, including to change its output.
* Share the system for others to use with or without modifications, for any
purpose.
These freedoms apply both to a fully functional system and to discrete elements
of a system. A precondition to exercising these freedoms is to have access to
the preferred form to make modifications to the system.
Preferred form to make modifications to machine-learning systems
The preferred form of making modifications to a machine-learning system must
include all the elements below:
* Data Information: Sufficiently detailed information about the data used to
train the system so that a skilled person can build a substantially equivalent
system. Data Information shall be made available under OSI-approved terms.
In particular, this must include: (1) the complete description of all data used
for training, including (if used) of unshareable data, disclosing the
provenance of the data, its scope and characteristics, how the data was
obtained and selected, the labeling procedures, and data processing and
filtering methodologies; (2) a listing of all publicly available training data
and where to obtain it; and (3) a listing of all training data obtainable from
third parties and where to obtain it, including for fee.
* Code: The complete source code used to train and run the system. The Code
shall represent the full specification of how the data was processed and
filtered, and how the training was done. Code shall be made available under
OSI-approved licenses.
For example, if used, this must include code used for processing and filtering
data, code used for training including arguments and settings used, validation
and testing, supporting libraries like tokenizers and hyperparameters search
code, inference code, and model architecture.
* Parameters: The model parameters, such as weights or other configuration
settings. Parameters shall be made available under OSI-approved terms.
For example, this might include checkpoints from key intermediate stages of
training as well as the final optimizer state.
The licensing or other terms applied to these elements and to any combination
thereof may contain conditions that require any modified version to be released
under the same terms as the original.
Open Source models and Open Source weights
For machine learning systems,
* An AI model consists of the model architecture, model parameters (including
weights) and inference code for running the model.
* AI weights are the set of learned parameters that overlay the model
architecture to produce an output from a given input.
The preferred form to make modifications to machine learning systems also
applies to these individual components. Open-Source-models and
Open-Source-weights must include the data information and code used to derive
those parameters.
The Open Source AI Definition does not require a specific legal mechanism for
assuring that the model parameters are freely available to all. They may be
free by their nature or a license or other legal instrument may be required to
ensure their freedom. We expect this will become clearer over time, once the
legal system has had more opportunity to address Open Source AI systems.
Definitions
* AI system: An AI system is a machine-based system that, for explicit or
implicit objectives, infers, from the input it receives, how to generate
outputs such as predictions, content, recommendations, or decisions that can
influence physical or virtual environments. Different AI systems vary in their
levels of autonomy and adaptiveness after deployment.
* Machine learning: is a set of techniques that allows machines to improve
their performance and usually generate models in an automated manner through
exposure to training data, which can help identify patterns and regularities
rather than through explicit instructions from a human. The process of
improving a system?s performance using machine learning techniques is known as
training.
These freedoms are derived from the Free Software Definition.
https://www.gnu.org/philosophy/free-sw.en.html
Recommendation of the Council on Artificial Intelligence OECD/LEGAL/0449,
Organization for Economic and Co-operation Development (OECD), 2024
https://legalinstruments.oecd.org/en/instruments/OECD-LEGAL-0449
Explanatory memorandum on the updated OECD definition of an AI system, OECD
Artificial Intelligence Papers, No. 8, OECD Publishing, Paris
https://doi.org/10.1787/623da898-en
See FAQs https://hackmd.io/@opensourceinitiative/osaid-faq
Endorse the Open Source AI Definition https://opensource.org/ai/endorsements
Thank you for your interest in endorsing the OSAID. Being an endorser means
your name and organizational affiliation will be appended to list of supporters
of Version 1.0 of the Open Source AI Definition.
Your name
Your email
Your institution
Your role
Endorsement type
IndividualInstitutionalBoth
Your message (optional)
This form uses Akismet to reduce spam. Learn how your data is processed.
Join Us
------------------------------
Subject: Digest Footer
_______________________________________________
Link mailing list
[email protected]
https://mailman.anu.edu.au/mailman/listinfo/link
------------------------------
End of Link Digest, Vol 383, Issue 33
*************************************
