Inside the Russian processor Baikal M
In 2014, Baikal Electronics acquired a license for the most advanced processor core at that time. After a series of incidents that influenced the further outcome of events, in October 2019 the company officially presented to the general public the first Russian general-purpose processor implemented on the ARM architecture. The creators decide to name their unique domestic brainchild “Baikal-M”.
In October 2020, which embodied the work of about 50 engineers and three years of development, it was recognized by the Ministry of Industry and Trade as a second-level microcircuit produced in the territory of the Russian Federation.
In accordance with the current regulatory framework, the developer of a Russian second-level microcircuit must have the rights to its design documentation, and the design, development and testing of microcircuits must take place within the borders of our country. They differ fundamentally from level 1 microcircuits only in that their direct production is carried out abroad. The current state of affairs in Russia is such that not a single processor development of a small topology can receive the status of the first level by definition – we simply do not have the corresponding production facilities.
In 2017, Mikhail Svarichevsky performed the opening of the Baikal T microprocessor, a previous development of Baikal Electronics based on the MIPS architecture. Now we have a unique opportunity to take a full look at the fresh Baikal-M, one of the most significant processor developments in Russia in recent years.
So … Baikal M1000 is a general-purpose microprocessor manufactured according to 28 nanometer technological standards at the factories of the Taiwanese company TSMC using ten metal layers and including more than two billion transistors. The maximum TDP of Baikal M is 35 watts.
The processor is made in the BGA form factor with a substrate size of 40 by 40 millimeters. Each BGA pin is responsible for powering and functioning of a specific element on the chip. Baikal Electronics has a fairly detailed datasheet on the official website. By looking at the BGA pin map, you can easily determine the location of the functional blocks.
After removing the heat-distributing cover, we see a crystal with thermal paste applied to it.
The die area of the processor is 248.78 square millimeters. The size of the chip without the so-called “scribes lines” – areas around the perimeter of the chip, which are “rubbish” – 239.97 square millimeters.
When the future chips are already “printed” on the silicon wafer, it must be cut into separate component parts. There are special dividing paths between the topology of each crystal, along which cutting is carried out. After separation, the remnants of the dividing tracks remain on the crystal, their size is so small that manufacturers, when specifying the area of the microcircuit, often neglect these values.
The crystal, like most modern microprocessors, has contacts over the entire area, some of which are responsible for supplying power to the entire surface of the chip, some for high-speed and low-speed interfaces, etc.
If you bring the microcircuit closer to the lower left corner, you can see the logo of the Baikal company and the year of manufacture of the microcircuit.
After removing the metallization, we got to the most interesting thing – the structural elements.
Baikal M has 4 dual processor clusters with ARM Cortex-A57 (or ARMv8-A) cores, the first 64-bit HP cores from ARM, clocked at up to 1.5 GHz. Each core has a 512 kilobyte L2 cache and a L1 cache that includes 32 kilobytes of data and 48 kilobytes of instructions. In the upper right part of the crystal, the Mali-T628 graphics processor is integrated with eight shader cores (in fact, two quad-core clusters), using a separate shared L2 cache of 256 kilobytes (128 kilobytes for each cluster). The shader core is comparable to AMD / NVidia’s shader engine and GPUs. It includes geometry, pixel pipelines, TMUs, and vectors with a scalar engine. The GPU cluster itself operates at up to 700 MHz.
Along the entire perimeter of the chip, there are monitoring units, which we can admire a little closer.
In the central part of the chip is the CoreLink Cache Coherent Network (CCN), which is used with eight megabytes of system L3 cache. Baikal M uses CCN 504, which can accommodate up to four processor clusters (each with 4 cores) and maximum support for up to sixteen megabytes of L3 cache. CCN organizes the coherence of all interfaces on a chip, a kind of Infinity Fabric from AMD.
Two memory controllers operating in 72-bit mode (8 bits of which are ECC memory) support DDR3-1600 and DDR4-2400 standards.
In the lower right part of the chip there is a complex of 16 PCI-e 3.0 lanes together with the controller, a control system for various interfaces.
In the upper right part of the chip are the rest of the peripheral units: two ten-gigabit Ethernet interfaces, as well as USB 2.0 and 3.0, SATA interfaces with their accompanying controllers.
To the left of the system cache between the two processor clusters is the HDMI / LVDS hardware video decoding unit. This interface allows video playback at 2560 x 1440 @ 60Hz.
Hmmm … a processor is a tricky thing. Really on the edge of the possible. It is useful to remember when, in the heat of another dispute, there is a desire to call the employees of Baikal or MCST idiots. Think what kind of intellectual work is behind the development of not even the fastest processor.
The semiconductor industry is the most complex infrastructure in the world. We used to think that space is the most complex industry, but one country can independently make a space rocket, but a modern chip cannot. Now at TSMC there are the most high-tech semiconductor lithography steppers in the world already using 5 nanometer technology, and everything goes to the fact that 3 and even 2 nanometers will also be available. Among other things, one such factory has a whole network of suppliers of resources required for the production of microcircuits. Plus there is specialized software, robots, highly qualified specialists. There is the Dutch ASML, there is the American Applied Materials, there is the French Air Liquide, the Japanese manufacturers of photoresists … And all this comes together in a complex huge production. With this in mind, one can answer the question: “Is technological autonomy possible?” No, it’s impossible. It is impossible to make completely separate production within the country.
In the 60s, each company had to provide its own production. Then software developers appeared, manufacturers of IP blocks emerged. Gradually, the number of factories began to decline. Each next stage became orders of magnitude more expensive than the previous one. Among other things, TSMC’s investment of seven nanometers is measured in tens of billions of dollars. And everyone understands that it is possible to recoup an investment only with a huge volume, that everyone needs to do this volume in one place and then get the best price.
There are long-lived technologies in the world, for example, 90 nanometers. Many companies are deploying production on old technological processes, since within the framework of the production in which they are engaged, the use of extremely delicate technological standards is impractical in many respects.
If we understand that we will produce hundreds of thousands of wafers per month at this factory (and hundreds of thousands of seven nanometer wafers are millions or even tens of millions of chips), if we understand who will come to us to produce there, probably about it makes sense to say it. You will not send seven nanometers into space, seven nanometers you will not put in a car, and seven nanometers you will not use in household appliances. Seven nanometers are needed where the lowest power consumption and extremely high performance are needed.
Photos of the crystal of the Baikal M processor, taken Fritzchens Fritz in full quality:
Thanks to Locuza for the identification of functional blocks on the chip.
Thank you for the attention!