©Mark Ollig
It is exceptionally larger in both physical size and processing capability than any other single computing chip used today.
Manufacturing of this chip requires a unique wafer-scale integration process used by Cerebras, a company located in Los Altos, CA.
A wafer (also called a substrate) is a thin, circular slice of silicon or other semiconducting material about 12 inches in diameter.
Computer chip components, such as memory or processors, are made using thin circular disks of silicon. One silicon disk is called a wafer. About 100 computer chips can be made on a single wafer.
Cerebras explained its unique wafer-scale integration process as making a single computing processor chip using most of the silicon wafer.
One Cerebras WSE (Wafer Scale Engine) computing chip measures nearly 8.5 inches by 8.5 inches, or slightly more than 72 square inches, which is about the size of your average dinner plate.
Each WSE chip holds an incredible 1.2 trillion transistors – yes, trillion.
By comparison, the AMD (Advanced Micro Devices) central processing unit chip measures 1.27 square inches and contains 32 billion transistors.
In 1984, placing one million transistors on a chip was reached, using ultra-large-scale integration.
Silicon chips are manufactured in chip fabrication facilities using methods much like a newspaper printing press.
The newspaper printing process uses ink; the silicon chip printing process uses photolithography, chemical deposition, and etching.
The chip fabrication facilities are located in super-clean rooms, where the folks wear those special white suits and head garb.
Let us compare the number of transistors in a WSE chip (1.2 trillion) with the amount used in the pocket-sized portable transistor radios we carried around with us back in the day.
Those small transistor radios contained up to eight electronic transistor components, used for amplifying or the switching of electronic signals.
Of course, today’s transistors are nanoscale-sized, and can only be seen using extremely powerful microscopes.
Before transistors, radios used vacuum tubes; but that is a topic for a future column.
In the digital computing world, transistors are mostly made of silicon or other semiconductor material.
They are used as a binary on-and-off switch to allow, or prevent current flow.
Transistors wired in specific combinations are the “logic gates” used in digital computing circuitry.
The WSE chip contains 400,000 programmable computing cores used for independent mathematical operations. These cores are small, fast, and include no volatile memory caches – think old-school RAM (Random Access Memory).
Cerebras said its WSE chip could be used in AI (artificial intelligence) research for neural networks requiring tremendous computing speed, memory, and communication bandwidth.
When installed, the WSE computing chip consumes a lot of power.
Heat radiated from the electronic transistors, and other components on the chip must be removed to keep the computing chip working.
This heat cannot be removed using conventional airflow cooling methods that are used with smaller wafer computing chips.
The larger wafers used with the WSE chip are manufactured with a “sandwiched cold plate” on top of the silicon.
Heat is removed by keeping the wafer’s silicon surface in physical contact with a metal manifold plate, which houses a vertically-mounted chilled-liquid cooling system.
With the heat removed from the wafer, the computing chips’ electronic components can operate at optimum efficiency.
The WSE chip will be used in commercial data centers and will assist with self-driving automobiles, gaming graphics, and artificial intelligence neural network research.
The computing chip is not available to the consumer market at this time.
Cerebras is to begin shipping orders to customers this month. They have not made public the cost of their WSE processing chip.
Dec. 23, 1947, a working transistor was demonstrated by scientists John Bardeen, Walter Brattain, and William Shockley inside Bell Laboratories in Murray Hill, NJ.
July 8, 1949, a patent was filed for a semiconductor amplifying device (hearing aid) with five transistors on a single substrate material by Dr. Phil Werner Jacobi.
The Cerebras website is located at https://www.cerebras.net.
It is exceptionally larger in both physical size and processing capability than any other single computing chip used today.
Manufacturing of this chip requires a unique wafer-scale integration process used by Cerebras, a company located in Los Altos, CA.
A wafer (also called a substrate) is a thin, circular slice of silicon or other semiconducting material about 12 inches in diameter.
Computer chip components, such as memory or processors, are made using thin circular disks of silicon. One silicon disk is called a wafer. About 100 computer chips can be made on a single wafer.
Cerebras explained its unique wafer-scale integration process as making a single computing processor chip using most of the silicon wafer.
One Cerebras WSE (Wafer Scale Engine) computing chip measures nearly 8.5 inches by 8.5 inches, or slightly more than 72 square inches, which is about the size of your average dinner plate.
Each WSE chip holds an incredible 1.2 trillion transistors – yes, trillion.
By comparison, the AMD (Advanced Micro Devices) central processing unit chip measures 1.27 square inches and contains 32 billion transistors.
In 1984, placing one million transistors on a chip was reached, using ultra-large-scale integration.
Silicon chips are manufactured in chip fabrication facilities using methods much like a newspaper printing press.
The newspaper printing process uses ink; the silicon chip printing process uses photolithography, chemical deposition, and etching.
The chip fabrication facilities are located in super-clean rooms, where the folks wear those special white suits and head garb.
Let us compare the number of transistors in a WSE chip (1.2 trillion) with the amount used in the pocket-sized portable transistor radios we carried around with us back in the day.
Those small transistor radios contained up to eight electronic transistor components, used for amplifying or the switching of electronic signals.
Of course, today’s transistors are nanoscale-sized, and can only be seen using extremely powerful microscopes.
Before transistors, radios used vacuum tubes; but that is a topic for a future column.
In the digital computing world, transistors are mostly made of silicon or other semiconductor material.
They are used as a binary on-and-off switch to allow, or prevent current flow.
Transistors wired in specific combinations are the “logic gates” used in digital computing circuitry.
The WSE chip contains 400,000 programmable computing cores used for independent mathematical operations. These cores are small, fast, and include no volatile memory caches – think old-school RAM (Random Access Memory).
Cerebras said its WSE chip could be used in AI (artificial intelligence) research for neural networks requiring tremendous computing speed, memory, and communication bandwidth.
When installed, the WSE computing chip consumes a lot of power.
Heat radiated from the electronic transistors, and other components on the chip must be removed to keep the computing chip working.
This heat cannot be removed using conventional airflow cooling methods that are used with smaller wafer computing chips.
The larger wafers used with the WSE chip are manufactured with a “sandwiched cold plate” on top of the silicon.
Heat is removed by keeping the wafer’s silicon surface in physical contact with a metal manifold plate, which houses a vertically-mounted chilled-liquid cooling system.
With the heat removed from the wafer, the computing chips’ electronic components can operate at optimum efficiency.
The WSE chip will be used in commercial data centers and will assist with self-driving automobiles, gaming graphics, and artificial intelligence neural network research.
The computing chip is not available to the consumer market at this time.
Cerebras is to begin shipping orders to customers this month. They have not made public the cost of their WSE processing chip.
Dec. 23, 1947, a working transistor was demonstrated by scientists John Bardeen, Walter Brattain, and William Shockley inside Bell Laboratories in Murray Hill, NJ.
July 8, 1949, a patent was filed for a semiconductor amplifying device (hearing aid) with five transistors on a single substrate material by Dr. Phil Werner Jacobi.
The Cerebras website is located at https://www.cerebras.net.
 |
| The Cerebras WSE computing chip compared with a baseball Photo used with permission |
 |
| The Cerebras WSE computing ship compared with a standard computing chip Photo used with permission |
