China claims record for overclocking memory chips without access to advanced lithography

Chinese memory chip manufacturer CXMT announced a "new breakthrough." DDR5 modules based on its Colorful brand chips were overclocked to a record 8800 MT/s (millions of data transfers per second) on an AMD X870E motherboard.

The achievement was confirmed in two popular configurations: 2x16GB and 2x24GB. The modules also passed internal stability tests. Just a few weeks ago, the same memory was overclocked to 8600 MT/s—a 200 MT/s improvement achieved in a very short time.

It's telling that the Chinese company achieves such results using equipment without access to advanced EUV lithography due to U.S. export restrictions. The company is forced to use older 16-nm EUV machines. Nevertheless, its products are already approaching the speed of the best SK hynix chips and perform well on both AMD and Intel platforms.

In recent years, competition in the global DRAM chip market has been intensifying. The traditional "big three"—Samsung, SK Hynix, and Micron—still control nearly 90% of the market. However, CXMT, founded ten years ago and only starting DDR5 production in late 2025, has already become the fourth-largest DRAM manufacturer in the world, with a share of approximately 7,6%.

For China, CXMT's achievements are of strategic importance. Beijing views the development of its own semiconductor manufacturing industry as critical to achieving technological independence. Plans to build new factories are intended to strengthen the country's position in the race for chip self-sufficiency. China aims to capture 30-35% of the global memory market by 2030.

Our country's share of global semiconductor production (in total) is no more than 0,3%. This component certainly needs to be increased to avoid ultimately becoming totally dependent on foreign suppliers, who might supply memory chips today, but tomorrow suddenly announce the impossibility of such exports, for example, due to sanctions, as has happened more than once in recent history. stories.

Comments

[rocket757]

Who seeks will always find!!!
The prerogative of creating and using high technologies does not belong solely to the countries of the alliance that the striped whales have assembled, around themselves and for themselves...

[ASG7]

Optimism is good, but thinner lithography allows for lower voltage and higher frequency due to the sensitivity of the elements, not just increasing the number of parallel flow processing algorithms. It's like building a 100-meter steam locomotive rather than a diesel locomotive.

[spektr9]

Optimism is good, but thinner lithography allows for lower voltage and higher frequency due to the sensitivity of the elements.

This is certainly very wonderful and interesting, but for modern AI clusters and data centers, maximum memory overclocking is not used because the stability of the modules is a priority, and considering that Chinese DDR5 is already widely used by HP, Dell, Acer and Asus, the product turned out to be clearly competitive

[ASG7]

We speak different languages. It's not the money that makes the technology, but the technology that makes the money.

[spektr9]

We speak different languages

Why are you trying to make a product that's already being used by the world's largest companies seem uncompetitive? Trying to portray the thinner lithography of Western modules as a huge advantage, when in fact it's nothing more or less than that...

[ASG7]

I wanted to say that China has no future without the development of lithography, yes, today it will do and bring in money, but the future belongs to technology (this is also relevant for us), and I didn’t say a word about the West.

[spektr9]

I wanted to say that China has no future without the development of lithography.

It's funny to read such statements after seeing how the PRC is rapidly developing its own national lithographs...

[rocket757]

Let's assume that we just make it smaller, frequency, voltage, etc.... there is no direct relationship here...
If we delve into the details, we'll have to remember the materials used in the production process and... there's a lot more to remember and take into account.
All this is top-notch work, the most modern technological processes...
And yet, this isn't the point... considering the head start that the major Western companies have, there are also achievements by others outside their circle...
Time will tell what happens next...

[Vadim S]

I'm afraid this isn't possible here! We haven't fallen behind, it's just in its infancy, and it's hard to say what it will develop into. Someone might say it's easier to buy from China, why make your own, as usual?

[opuonmed]

Quote: Vadim S

Someone might say that it's easier to buy from China, why make your own, as usual?

This is said by those who want to see Russia as a user (controlled and not sovereign), rather than a creator (creating technology and being sovereign in it). This is the only way to explain the statements of those who say it's better to buy from abroad than to create and make your own.

[Lionnvrsk]

Quote: opuonmed

This is said by those who want to see the Russian Federation in the user's camp (which is controlled and not sovereign)

Not long ago, this was practically a government strategy: "Why bother? We have plenty of oil dollars and gas. We just snap our fingers and they'll bring it all to us."

[Orange Bigg]

Quote: opuonmed

Quote: Vadim S

Someone might say that it's easier to buy from China, why make your own, as usual?

This is said by those who want to see Russia as a user (controlled and not sovereign), rather than a creator (creating technology and being sovereign in it). This is the only way to explain the statements of those who say it's better to buy from abroad than to create and make your own.

And yet, despite everything, good news about lithography is coming. Information is already in the media.

Russian developers have created and tested a multicathode—a single source of multiple electron beams—for a promising lithograph, allowing for silicon wafer exposure 3000 times faster. The first prototype of the machine itself is expected around 2029; the developed technology covers standards from 350 nm down to single nanometers.
Russian developers, with support from the National Technology Initiative (NTI), have created and tested a key technology for multi-beam electron lithography: a multicathode, a single source that simultaneously generates multiple electron beams instead of just one. Unlike systems that rely on complex beam splitting with numerous mirrors and optical shutters, this development creates independent electron sources on a single substrate, reducing infrastructure requirements, power consumption, and equipment costs. The technology developed by Russian engineers is designed for the production of microchips with topological standards ranging from the traditional "massive" 350 nm to promising processes measuring just a few nanometers. The ultimate resolution is determined by the characteristics of the photoresists used, not the capabilities of the beam source.

https://tehnoomsk.ru/archives/27078

[spektr9]

And yet, despite everything, good news about lithography is coming. Information is already in the media.

Information is very good, but a dead industry is bad, and so far there have been no significant changes in the industry other than the Russian-Belarusian lithographer and promises to marry.

[Orange Bigg]

The above information speaks of a simpler technology for creating lithographs, without numerous lenses and chemicals. The Russian-Belarusian lithograph has been created at 350 nm, and plans are underway for 130 nm and 90 nm, but beyond that, everything is complex and time-consuming. So, it's simpler to follow the path outlined in my previous post. In principle, 2029 is just around the corner.

... In 2026, work will begin on creating a Russian lithograph for producing chips using a 90 nm topology. ... in November 2026, he plans to complete the development of a stepper with a resolution of 130 nm, he noted. ... "Developing a full-fledged lithograph from scratch can take 5-10 years and require hundreds of millions of dollars," Dmitry Pshichenko, an expert at the Russian Scientific Research Institute of Microelectronics and a member of the International Academy of Telecommunications, told CNews. "However, if we are talking about localizing, adapting, or refining existing solutions (for example, based on 130 nm), the timeframe and costs can be significantly lower: about 2-4 years." ... Once the development of a 130 nm lithograph is mastered, there is no need to redesign all the equipment. It is necessary to refine the existing elements and technologies (replace lenses, laser modules, and the control system) in order to squeeze a resolution of 90 nm out of the same platform.

https://www.cnews.ru/news/top/2026-03-23_v_2026_godu_rossiya_nachnet_rabotat

[spektr9]

So it's easier to follow the path outlined in my previous post. Basically, 2029 is just around the corner.

This, excuse me, reminds me of the statement about a lunar station by 2015.

[Themistocles]

Already in total dependence... And regarding China... this is what life-giving Western investments do. The result, as they say

[Lt. Air Force stock]

The limit of process technology reduction will soon be reached. Next up: neuromorphic, photonic, and quantum (in limited areas of application) processors. It makes more sense for us not to try to reach 0,5-0,1 angstroms by investing huge amounts of money, but to develop more promising areas.

[alexoff]

Quote: Lt. air force reserve

It makes more sense for us not to try to reach 0,5-0,1 angstroms.

The size of a silicon atom is greater than one angstrom; there are no technologies, nor will there ever be ones, that are a fraction of an angstrom.

[stelltok]

It makes more sense for us not to try to reach 0,5-0,1 angstroms by investing huge amounts of money, but to develop more promising areas.

What directions?

[Orange Bigg]

Quote: Lt. Air Force stock

The limit of process technology reduction will soon be reached. Next up: neuromorphic, photonic, and quantum (in limited areas of application) processors. It makes more sense for us not to try to reach 0,5-0,1 angstroms by investing huge amounts of money, but to develop more promising areas.

Well, that requires technology. But it's already entirely feasible to create a domestic quantum photonic computer in a laboratory, and Russian scientists have succeeded. However, for it to become a mass-market tool, many engineering and software challenges still need to be addressed. So, it's safe to say: the foundation has been laid, and progress is underway.

[Normann]

Our country's share of global semiconductor production (in total) is no more than 0,3%. This component certainly needs to be expanded.

and before this it wasn’t clear that we depended on them?

[novel66]

Build up?? They can't or don't want to build the AN-2. Oil and construction are our everything.

[Lt. Air Force stock]

Quote: novel xnumx

Build up?? They can't or don't want to build the AN-2. Oil and construction are our everything.

The problem with an An-2 analogue is making it cheap, which is difficult to achieve with small-scale production. The required performance figures can't be achieved solely through the domestic market, and no one will buy it abroad.

[Yes it's me]

If you make a perfect device that is an order of magnitude superior to its analogues, then there will be...

[alexoff]

The problem with the An-2 replacement is that it was given to hacks, who had the plane designed by MAI students, and the factory pretended everything was fine—after all, money was being allocated! Now Badekha heads the UAC; the state has recognized his ability to organize the production of light aircraft, and apparently wants to eliminate the need for heavy aircraft.

[dementor873]

That's how everything is built here, with profit and profit taken into account, and the need to provide remote areas with transport accessibility doesn't fit into a market economy.

[Motorist]

Quote: novel xnumx

They can't build the AN-2.

It's unclear whether the air conditioning is working or not, but they're trying to build a bridge! (Film Election Day).

[Rockerman]

Overclocking is fine, but what about the lifespan? I wouldn't buy modules like that.

[smart fellow]

Apple is seeking US government approval to purchase RAM chips from the Chinese company CXMT. They must be idiots at the top. Why don't they consult with you?

[eon eon]

They will buy according to strict internal quality criteria, the rest will be sold at retail... Do our users really need such good stuff?

[smart fellow]

Major memory manufacturers, such as Samsung and the Chinese company CXMT, do not sell defective batches as normal. If defects are high, problematic silicon wafers (packages) are disposed of, recycled, or rejected during the testing stage, reducing the final chip performance.
General standards for handling defective goods
Testing (Sorting): Each crystal is tested for functionality and operating frequencies.
Grading: Partially defective chips are sold as slower or stripped-down versions (e.g., with less memory or fewer channels). They are marked accordingly.
Disposal: Unusable crystals are sent for crushing and extraction of valuable chemical elements or silicon.

In the 90s, there was a company in Russia called "Land" that bought batches of computer components and monitors with a high defect rate. They assembled computers from them and sold them at reduced prices in Russia. They were given 10 days for testing, after which the buyer could simply return the goods without explanation.

So, this isn't the manufacturer's problem, but the consumer's. Who's to blame for Americans having higher quality standards? It's the greed of Russian traders, not deception by Chinese manufacturers.

In China, sellers are required by law to notify buyers of defects and substandard goods. This is enshrined in the Consumer Protection Law of the People's Republic of China (Article 23) and in the online retail regulations, which require sellers to clearly indicate any discounts or defects before the purchase is made.
What does Chinese law say?
Right to information: The buyer has the right to know the actual condition of the product, its properties and possible defects.
Defect Exclusion: If the seller has provided the buyer with advance and fair notice of specific defects (minor flaws) prior to purchase, the buyer will no longer be able to claim compensation or a refund for these specific defects.
Liability for concealment: If the seller conceals a defect or substandard product, the product is subject to return, replacement, or repair at the store's expense. Concealment of dangerous defects will result in severe penalties.

[naofumi]

Overclocking is fine, but what about the lifespan? I wouldn't buy modules like that.

Colorful's full specifications were carefully blurred out in the screenshots, but for 8800 MT/s using a Chinese process technology, they've 100% loaded them with more than 1.5 V. These modules will last as long as a hamster in a microwave. Buying these for home use is like taking out a loan for a one-way honeymoon to Chernobyl. But for a few minutes, you'll be the fastest guy in the hardware cemetery.

[Sensor]

I remember 10-15 years ago, a small company advertised computers with clock rates several times higher than those of high-end desktop computers. The secret was simple: all the hardware was immersed in a circulating liquid that provided heat dissipation. All components were overclocked to their maximum. However, this didn't prevent localized overheating within the microchips, and they didn't last long.

[opuonmed]

However, CXMT, which was founded ten years ago and only began producing DDR5 in late 2025, has already become the fourth-largest DRAM manufacturer in the world with a share of about 7,6%.

Well done!

Due to export restrictions, the company is forced to use older 16-nm DUV installations. Just a few weeks ago, the same memory was overclocked to 8600 MT/s—an improvement of 200 MT/s was achieved in a very short time.

That's also good! DDR6 is coming soon, and it will have a base data rate of 8800 Mbps.

Our country's share in global semiconductor production (in total) is no more than 0,3%

And this is complete rubbish, he can’t even provide for himself yet!

[topol717]

Memory is made of capacitors, not transistors. A delicate manufacturing process isn't particularly necessary, since capacitor capacitance is dependent on surface area. As soon as this sector became profitable, China jumped on it and is probably not going away. But there's another important factor: Samsung memory, for example, consumes less power and runs much cooler than similar Xenix memory. So speed is far from the most important factor.

[smart fellow]

In random-access memory (DRAM), a capacitor stores a bit, and a transistor acts as a switch. They both work together. The capacitor holds a charge (1 or 0), and the transistor allows access to it.
Static RAM (SRAM): Made of only transistors (flip-flops) without capacitors.
Flash memory (SSD): Uses special floating-gate transistors that hold a charge for years without power.

There is a delicate technical process there and it is not particularly necessary

Server-grade DRAM modules have capacities of up to 512 GB, and SSDs (enterprise-class drives from Kioxia) have capacities of 245,76 TB. Without a reduction in process technology (currently 4 nm), achieving such capacities is impossible without using multilayer packaging, which is also a delicate process technology—for example, printing 400 layers of transistors for an SSD.

[jovanni]

China claims record for overclocking memory chips without access to advanced lithography

Yeah, China, China can do it! But we don't seem to have any intentions...

[vromi]

What can our country boast about?

[Skobaristan]

Yours has nothing to do with it. After all, judging by the wording of the question, the Russian Federation is not your country. Moreover, bragging is a strange trait, especially at the state level. But yes, Russia possesses advanced nuclear technology. And this is one of the most high-tech and complex sectors of science and technology.

[gridasov]

It's strange that no one is pursuing the goal of creating a technology for working with big data that would eliminate the very concept of data transfer speed. This is being achieved through the development of a mathematical methodology for distributing extremely large data sets and, of course, a hardware method based on expanding the functionality of the device that transforms electromagnetic pulses into units of information code.

[tjeck91]

I used to be a PC enthusiast, so this news is meaningless. Such overclocks are performed under nitrogen, and the runtime of such a module is measured in minutes, tens of minutes at most. In other words, this record has no practical meaning. It's simply a measurement of the numbers and two particularly well-chosen memory modules.

[Rockerman]

Quote: smart fellow

Apple is seeking US government approval to purchase RAM chips from the Chinese company CXMT. They must be idiots at the top. Why don't they consult with you?

You apparently didn't read my comment carefully. Well, since you're so open about it, please confirm that Apple will be purchasing exactly these overclocked memory modules. Otherwise, your comment is just a standard attack on a known substance.

[solovjoff]

Our country's leadership is completely uninterested in this area. It's just that... This shortsightedness (declaring cybernetics, psychology, and the like pseudosciences) has been going on since the Soviet era. It looks like we'll remain resource suppliers to the developed world.