Russian Electronics: Semiconductor and Processors

Every server begins from its chassis and some general features such as remote management, which Sber evaluates under its Functional Testing procedure. Apparently, an MCST Elbrus-8C machine failed 84% of Sber's Functional Testing as it could not be easily removed from the rack, lacked proper LED indicators, and came without remote management, which to a large degree made it unusable for usage in commercial datacenters. There is some hope, though.

"One of the surprising things about the Elbrus-8C server was that it is a real product," said Zhbankov. "It was a real server that we were given. […] It is an actual product that has its disadvantages, loads of disadvantages, but we can work with them."

Elbrus-8C Evaluation Summary
4-way Elbrus-8C vs 2-way Intel Xeon Gold 6230
SPEC CPU 2017 2.62 (base) ~ 3.15 (peak) times slower
PGbench/PostreSQL 1.7 (read only profile) ~ 3.3 (read write profile) times lower
Java 23 ~ 26 times higher response time

lancelot wrote:

To be honest I expected worse given it is a 28nm vs 14nm processor.
Given Elbrus is VLIW, and not that suited for multi-threaded workloads like a bank would use, but more for peak FP compute I think it did ok.

Sber basically has an issue with the lack of suitability of the server they were provided with for rack mount and remote service. That is an issue of the platform not the CPU design itself. The poor Java performance might be a lot of things like having less and slower memory or the low amount of threads on each core.

I think a bank would be better off with the Baikal line of processors though. Short and small core with multiple threads is better than Wide and large core with wide FP units for a bank running SQL and Java loads.

miketheterrible wrote:Sorry, had to resort to a Twitter post:

Baikal-S, a 48 arm processor that also uses RISC-V coprocessor.

https://mobile.twitter.com/torgeek/status/1472571292360626186

Singular_Transform wrote:

lancelot wrote:

To be honest I expected worse given it is a 28nm vs 14nm processor.
Given Elbrus is VLIW, and not that suited for multi-threaded workloads like a bank would use, but more for peak FP compute I think it did ok.

Sber basically has an issue with the lack of suitability of the server they were provided with for rack mount and remote service. That is an issue of the platform not the CPU design itself. The poor Java performance might be a lot of things like having less and slower memory or the low amount of threads on each core.

I think a bank would be better off with the Baikal line of processors though. Short and small core with multiple threads is better than Wide and large core with wide FP units for a bank running SQL and Java loads.

It is a good news, they started to push the CPU to wider application, like datacenters.

Main problem here is the Elbrus needs new linux kernel and application library, it is not a native x86 CPU, needs to translate the instructions.


And these tests compared a 20 core CPU again an 8 core one, means the results are the expected one. They need to put two CPU into each server, and even that will be cheaper, considering the Elbrus die size fraction of the Xeon.

Main problem here is the Elbrus needs new linux kernel and application library, it is not a native x86 CPU, needs to translate the instructions.

Russian scientists have created the first domestic five-qubit integrated circuit for quantum computing. MIPT specialists worked on it, and it is a full-fledged Russian prototype of a quantum processor that can be used in quantum machine learning. At the beginning of 2020, IBM CEO Arvind Krishna openly expressed doubts that Russia is capable of making a breakthrough in the field of quantum computing.

Domestic quantum processor

Russia has developed the first integrated circuit based on five superconducting qubits in a holder. It was created by specialists of the Moscow Institute of Physics and Technology (MIPT) in the Laboratory of Artificial Quantum Systems (LIKS), and, as representatives of the university told CNews, this development can be considered a prototype of a quantum processor.

The developers of this multi-qubit system claim in their official announcement that it is unique and completely manageable. According to them, even at the current stage of development, it can be used in quantum machine learning — a separate field of science at the intersection of quantum physics and modern information processing technologies.

Created in the MIPT laboratory, the integrated circuit was manufactured with the participation of employees of the Center for Collective Use (CCP). At the time of publication of the material, it passed a number of tests that showed that all its elements work exactly with the parameters that the developers expected.

Six years of operation

The first Russian qubit, according to a LIKS researcherAccording to Alexey Bolgar, it was obtained six years ago, in 2015, directly in this laboratory. According to him, after that, the laboratory staff and the CCP continued to work in this direction. "All these years, the staff of the MIPT Central Design Bureau and the laboratory have been working to improve the technology of manufacturing superconducting quantum structures with different architectures. As a result, we now have a technology that is already reliable enough to create multi-qubit computing devices. The integrated quantum circuit created by us, unlike the prototypes previously developed in Russia, allows us to fully control the state of all five qubits. Such integrated circuits are necessary for creating a universal quantum computer based on superconducting qubits. This is a great technological success, " said Alexey Bolgar.

MIPT representatives noted that the creation of the Russian multi-qubit integrated circuit was made possible due to four factors, and the first of them is a significant improvement in the control of geometric and electrical parameters of tunnel contacts. According to representatives of the university, these contacts can be considered the "heart" of superconducting qubits, since the performance of the entire quantum circuit directly depends on the quality and reproducibility of their manufacture.

The second factor is the adjustment of the manufacturing technology of microwave resonators, the quality factor of which in the single-photon mode is hundreds of thousands. This is also a very important part of quantum integrated circuits — they are needed to read the quantum state of qubits.

The third factor is debugging the manufacturing process of" hanging bridges " (air bridge), which are necessary to suppress parasitic resonant modes, which has a positive effect on the Q-factor of structures. But the most important component that allowed MIPT specialists to create a multi-qubit system, in their opinion, is the experience they have accumulated in this area over the past few years.

Future plans

MIPT does not specify when exactly the era of Russian quantum computers will begin, nor does it disclose its future plans for the development of new multi-qubit integrated circuits and their implementation. According to Alexey Bolgar, who was directly involved in the development of the five-qubit scheme, for any further actions in this area, it is necessary to modernize both the CCP and the ICS laboratory as part of MIPT.

"Our current results show that the technological and measurement capabilities of the CCP and our laboratory allow us to work out and complete all the steps necessary to create elements of quantum processors, from technological drawings to an integrated quantum circuit on a chip and its measurements. However, further development of work on the creation of controlled elements of the quantum computer and the computer itself will require modernization of the "clean zone" of the CCP and additional equipment of the laboratory with modern research equipment, " said Alexey Bolgar.

Manufactured tunable ring resonator based on silicon nitride and chalcogenide phase memory material. General view and schematic image of the ring resonator obtained using an optical microscope (a), a scanning electron microscope (b), and a 3d illustrator (c).

Scientists of the Moscow Institute of Electronic Technology (MIET) and the Moscow State Pedagogical University (MSUU), together with other Russian scientists, have created a chip for developing new-generation photonic circuits. Experts say that the developed technology for manufacturing non-volatile tunable nanophoton chips is ready for implementation in microelectronic production without additional upgrades.

A team of scientists from MIET and Moscow State University has created fully optical tunable ring microresonators based on silicon nitride and thin films of one of the so — called phase memory materials-Ge-Sb-Te (GST). The main feature of these materials is the change in optical and electrical properties when switching between amorphous (disordered) and crystalline (ordered) states.

"In the chip we developed, the surface of silicon nitride ring microresonators is locally coated with a thin GST film. A change in the phase state of the GST coating and, consequently, its absorption leads to a change in the optical signal passing through the waveguide. Switching of phase states can be initiated by laser pulses passing through a waveguide, " said Petr Lazarenko, a senior researcher at the Institute of Advanced Materials and Technologies of the National Research University MIET.

Phase state switching occurs in 10 nanoseconds, and GST film is one of the most optimal materials for controlling signals in thin-film waveguide elements that are used in telecommunications devices.

According to the specialists of the research group, the technology they have developed for manufacturing non-volatile tunable nanophoton chips is fully ready for implementation in microelectronic production, since it can be fully implemented using standard CMOS technology processes and does not require additional modernization of installations.

Now the research team is optimizing the chip to increase the number of recorded logic levels. The research group also develops approaches to designing new integrated optical circuits and systems based on them.

A universal processor for performing computationally complex cryptographic transformations and, in particular, blockchain transactions with extreme energy efficiency.

Organized ten years ago at Moscow State University, the Malt System microelectronic design center developed and registered with Rospatent a processor with its own architecture for streaming network traffic processing.

Processors for energy-efficient solution of mathematical physics problems that require irregular memory access.

The Russian multi-core processor development team, Malt System, has passed the procedure for registering the topology of its new processor, having received a certificate from Rospatent.

In the company's model range, the novelty has the name Malt-Cv3, its commercial name is "Enceladus" (the name of one of the titans in ancient Greek mythology, which migrated to the sixth-largest satellite of Saturn).

According to the developers, Enceladus is designed for streaming network traffic processing at speeds up to 1 Gbit / s, including ensuring the security of network connections "by software encryption/decryption of traffic using any domestic or foreign algorithms."

The company calls the chip the third in a line of Malt-C processors designed to perform complex cryptographic transformations, and the first "embedded and extremely compact". The latter circumstance is explained by the fact that it will have dimensions of no more than 9×9 mm in a BGA-type case (English Ball Grid Array, array of balls). "This is less than the RJ-45 connector," the developers note. — Such a processor can really be integrated into almost any device."

Enceladus is designed using 16 nm technology. Malt System announced its creation in mid-May 2020. The company announced placing an order for the production of chips at TSMC's contract Taiwanese factory on July 30, 2021. Developers hope to receive samples "in silicon" in the first quarter of 2022. The release of the SDK and debugging board for Enceladus is scheduled for the third quarter of 2022.

Processors for parallel operation with large data sets that are stored in RAM or external memory and are characterized by complex processing logic.

Chip architecture and characteristics

The company notes that Enceladus is built on its own original Malt architecture. It is based on dozens or hundreds, depending on the model, of compact asynchronous universal computing cores connected by one or several original worm-hole networks with a fat-tree topology, the developers describe their architecture.

"Communication between networks is hardware and software," they say. - The hierarchy of universal cores includes three levels: supermaster — control core, master-communication cores, slave-computing cores available for user tasks. Slave cores can contain vector accelerators that perform specialized tasks of the target class. Each accelerator contains from eight to 128 processor elements of the same type with shared instruction memory. All computing cores and accelerators have their own local data memory. All generic cores directly address shared external dynamic DRAM memory and other shared resources (PCIe, Ethernet, SATA)."

Directly in Enceladus, four computing clusters are organized, each of which contains 16 specialized cores and is controlled by one universal RISC core. The chip contains two 1GEb Ethernet controllers, seven general-purpose RISC processor cores, three SIMD accelerator blocks, a shared static memory block, and SPI-Flash, UART, and GPIO controllers.

Interaction with the processor is carried out via universal interfaces SPI, UART, GPIO, which are also controlled by a single dedicated universal core. 512 KB of shared static SRAM memory will be available on the chip, and it is also planned to support up to 64 MB of external HyperRAM/HyperFlash memory.

The estimated operating frequency of the processor will be 1.2 GHz. The chip implements an IP frequency generator unit with automatic tuning without pulse interference FDPLL. The projected power consumption of the new product at full load should not exceed 3 watts. In typical modes, depending on the implemented algorithm, the power consumption should be in the range of 500-2000 MW.

Development status

MALT-Cv1-manufactured and tested samples in silicon.

MALT-Cv2-manufactured and tested samples in silicon.

MALT-Cv3 - currently in the RTL description debugging stage.

https://maltsystem.ru/ru/product/malt-processors#malt-d

miketheterrible wrote:Very good and informative video, thank you!

I look forward to future developments from MCST, Baikal and the makers of KOMDIV.

rigoletto wrote:https://www.youtube.com/watch?v=-ojbRZnJkJo&t=59s

The Russian military also characterized the mission as a "technological... equipped with newly developed instruments and systems for their testing under conditions of radiation and heavy particles."
...
According to radar tracking data from the 18th Space Control Squadron of the US Space Force, Kosmos-2553 entered a 1,987 by 1,995-kilometer orbit with an inclination 67.08 degrees toward the Equator and an orbital period of 126.99 minutes.

lancelot wrote:China has two sets of semi tool manufacturers. The ones who make tools for their MIC. The ones who make tools for commercial use.
The ones who make tools for the MIC are basically government research institutes, labs, and factories. They not export anything outside China.
Their tools are considered strategic materials and treated as such.

The ones who do the tools for commercial use sometimes export to the West themselves and might thus be vulnerable to Western sanctions.
But the commercial tools vendors quite often have limited exports to begin with because of Western competition so they might be amenable to a deal anyway.

China is also interested in breaking this dependence with the West themselves since companies like Huawei and Phytium are still in the black list.
Phytium is a company which made multi-core ARM based chips similar to Baikal.

I think Russia should expand 65nm/90nm production since the Chinese should already have 100% replicated this chain. That will take care of any critical semiconductors necessary for industrial control.

Russia can also likely secretly order 28nm production someplace since there are multiple fabs worldwide which have this technology.
This will make tracing the source more difficult. In the next 1-2 years the Chinese should have 100% replicated the 28nm tool chain too and then Russia can just build a fab using that.

Developing Russian tools is, I think, a waste of time for the most part.
What Russia should do is make an industry team to figure out which semiconductor materials and parts they need and come up with substitutes.

With regards to tools, Russia should work on making their own e-beam lithography technology analogous to the deal they had with Mapper Lithography before ASML bought it.
Russia (RUSNANO) still owns the facility which manufactured the e-beam write heads. The rest is basically a bunch of software and control computers.
https://www.eetimes.com/russia-backs-e-beam-lithography-firm/
https://semiengineering.com/what-happened-to-next-gen-lithography/
https://semiengineering.com/manufacturing-bits-feb-5/
https://mapperllc.ru/

If Russia gets e-beam technology working this would allow them to break the Western monopoly on both chip inspection and lithography.
E-beam is not suitable for mass production but for manufacturing just a couple hundred chips for the military and making custom designs it will work just great.

Whatever sanctions they put with regards to commercial semiconductor sales to Russia it should be relatively easy to source chips in the gray market.
Good luck tracing where second hand chips are sold to. As long as people do not mind using four year old hardware there should be more than enough available.

The rest might become really expensive and hard to source. But that is about it.

kvs wrote:There was open discussion of moving to 28 nm lithography in the last few years.   So I am rather confident that this was happening.
There was no reason to stall at 65 nm.

China is moving to 10 nm (real process resolution not the fake 7 nm claimed by TSMC) as well.   I have seen the usual spew from western
analists that China is "failing".   Their brains are failing and were never full of much other than shit anyway.

05.03.2022
Nanometer sanctions: an impossible war for microprocessors between Russia and the West

Opinions
OLEG IZUMRUDOV
Executive Director of the RosSHD Consortium.

The world's largest manufacturers of computer processors and semiconductors have announced that they will join the sanctions against Russia. Processors for user PCs, servers, and data storage systems will not be delivered to our country. Loud words inspire pessimism in the layman. In practice, such steps will leave almost the entire world without microelectronics.


Sine qua non

Talking about sanctions against Russian microelectronics is impossible without talking about global microelectronics. Threats to stop the supply of AMD and Intel to Russia, as well as stopping the production of Russian "Baikal" and "Elbrus" at the Taiwanese TSMC plant look like an information attack. The reality of implementing these plans in the conditions prevailing in the modern microprocessor industry is close to zero.

The fact is that our country today accounts for 80 percent of the market for sapphire substrates – thin plates made of artificial stone, which are used in opto-and microelectronics to build up layers of various materials, such as silicon. They are used in every processor in the world-AMD and Intel are no exceptions.

Our position is even stronger in the special chemistry of etching microchips using ultra-pure components. Russia accounts for 100 percent of the world's supply of various rare earth elements used for this purpose.

A ban on finished products for Russia will result in a retaliatory ban on the supply of production components and cause an acute shortage of processors for the whole world. In comparison, the situation with supply disruptions at the end of 2021 will seem relatively easy.

Such statements should only be taken as part of the information war – propaganda rarely takes into account the integration processes of the global economy.

Russia's integration into the global IT industry began in the heyday of the USSR, in the 1970s. Abruptly replacing us with the resource of alternative locations for the production of substrates will not work, and replacing rare earths from Russia will not work in any way.

There are objective reasons for this. To ensure the required quality of sapphire substrates, the plant needs to enter the so-called "extra clean mode". It involves about thirty years of continuous production with compliance with all the subtleties of the technological process. In addition, such plants can only be developed in conditions of almost zero seismic activity. The slightest fluctuation of the Earth's crust (1-2 points) will lead to the fact that absolutely all production will need to be restarted from scratch.

That is why the products of similar enterprises in California or Taiwan – and these are seismically active zones – are noticeably inferior in quality and volume to the level required in the industry.


Road to the East

But let's assume that the sanctioning party still has a trump card up its sleeve, and the threat of cutting off supplies from Russia is calculated. In this case, our country should also have a "just in case" scenario. And we have it.

It is known that in the last few months before 2022, extensive consultations were held on the possible exclusion of TSMC's Taiwanese plant from domestic production chains. It is an industry-leading manufacturing facility with approximately $ 48 billion in revenue and well-established 7-and 5-nm manufacturing processes and 3-and 2-nm chip production.

TSMC has approximately 54 percent of the global market, while the second closest competitor (Samsung) has 17 percent. The remaining volume is distributed among GlobalFoundries, Semiconductor Manufacturing International Corporation (SMIC), UMC and other niche players, although quite large in these niches.

So, the most obvious option, taking into account the specifics of the moment for Russia, is mainland China, where SMIC operates, and also develops the production of its own equipment for the production of processors at the AMEC factory, whose equipment allows processing 300-mm silicon substrates used for a wide range of technical processes from 65 nm to 5 nm. So far, the output of finished products is significantly lower than TSMC's resources, but it's only a matter of time.

This is not an easy path: the transition to AMEC will require a complex of works to adapt to the realities of a particular factory, changes to templates, documentation, and so on. However, this is not a radical change in the entire existing process, and when the plan is activated, it will take a year and a half.


Taiwan Question and Chinese Answer

However, you can "take a shortcut" and get access to TSMC's established production processes. To do this, China must resolve the Taiwan issue.

Beijing has never recognized the island's independence. The presence of an American military base in Taiwan does not stop him from wanting to reunite with the lost territory. However, in case of attempts to capture the island, the Taiwanese military is obliged to destroy TSMC with a missile strike. When this plan is implemented, the global microchip industry is coming to an end: Intel and Samsung production remain to cover the need for processors in the local US market.

China has been preparing to resolve the Taiwan issue for a long time, and the military scenario is far from the only one. Loyal political forces were nurtured on the island, business oriented towards the PRC developed – the economy of Taiwan has been seriously reoriented since the early 2000s, with an eye to the market of its older mainland brother. China can implement reunification within the framework of the "one language, one people" concept, taking into account the mass of kinship ties between the two regions. It is quite possible to carry out such integration from the point of view of ideology and propaganda.


Other regions

In addition to China, India has the status of a partner with huge potential in IT in general and microelectronics in particular.

After the end of the colonial model of cooperation with the British, India grew quite quickly and for some time actively moved towards the USSR. This made it possible to create a serious education system in this country, which absorbed all the best from the British and Soviet systems.

The country is interested in entering the global market as an independent player or a partner of one of the leading players, and it has all the necessary basic elements, especially the human resource, for this purpose. It is enough to pay attention to the number of Hindus in top management in IT giants and startups, including in the field of microelectronics. Synergy from cooperation with Russia in this area is very promising.

In general, today's agenda can be interpreted as a demonstration of the fundamental possibility of getting out of unequal financial relations with global leaders within the framework of the Western model of development. In these circumstances, an alliance with Russia may not only be quite realistic, but also very productive.


Third way: own unique technical process

In addition to purely geographical solutions to the problem, there is also a way to create a completely new technological approach to the production of semiconductors.

It consists in abandoning the use of silicon and replacing it with gallium arsenide. This is a semiconductor material from the class of compounds AIII BV-a dark gray crystal that is actively used in infrared optics, as well as in opto - and microelectronics. The fundamental difference from silicon is the ability to build three-dimensional chip architectures instead of two-dimensional wafer printing, which allows you to implement a different approach to creating processors.

The team developing the project was originally going to launch a startup in the UK. But nothing happened because of the pressure of Intel and AMD, which did not need such competition at the level of an alternative technical process. Upon my return to Russia in 2016, I managed to find application for the developments. Production of such solutions has been established today in Perm for the needs of the Ministry of Defense of the Russian Federation. However, the technology has all the possibilities for use in civilian tasks.

The advantage of this development scenario is that there are simply no analogues of this production in the world. This is your own market: "blue ocean" multiplied by "green field". The only question is state support and political will.

Research on the civilian potential of gallium arsenide is being conducted quite intensively. Of course, you don't have to wait for an instant result. But such a path can become an element of a combined approach to solving the problem: traditional silicon solutions based on a two-dimensional technological process based on cooperation with China, plus gallium arsenide know-how.