During the IEEE International Electron Devices Meeting (or IEDM (open in new tab)), Intel claimed that circuits will reach a trillion transistors by 2030.

At the meeting, Intel's Component Research Group presented its predictions for the future of circuit manufacturing (via sweclockers (opens in new tab)) and how new packaging technologies and materials will enable chipmakers to manufacture chips with 10 times the transistor density while maintaining Moore's Law. described how they will enable chipmakers to manufacture chips with 10 times the transistor density while maintaining Moore's Law.

Moore's Law (opens in new tab) is the principle that computer speed and capability doubles every two years as the number of transistors that fit on a microchip increases. If my calculations are somewhat accurate, it is not so far-fetched to achieve one trillion transistors in eight years if the industry manages to double the current number every two years until then.

Given the limited space on chips, it is possible but not easy. According to AMD's Mark Papermaster in a recent Q&A (open in new tab), "doubling the transistor density every 18-24 months is possible, but no longer possible in the same cost range."

Transistors act like small switches inside the processor that allow current to flow through the chip. Depending on the contents of the chip, it may have millions or billions of transistors. Currently, the maximum number of transistors in a commercial processor is 114 billion. This clear honor is given to Apple's M1 Ultra chip (open in new tab) in the Mac Studio desktop computer; Nvidia's RTX 4090 (open in new tab) GPU has about 76 billion transistors, Intel's Ponte Vecchio supercomputing GPU has about 100 billion.

But how does Intel plan to reach 1 trillion transistors? The company is moving away from monolithic circuit design and adopting a chiplet design like its rival AMD. This chiplet design allows Intel to break it up into several circuits, giving it more space and cost efficiency to add transistors via discrete dies.

Intel's first chiplet design, the Meteor Lake CPU due out next year, leverages Intel's Foveros stacking technology, which allows the Arc GPU (opens in new tab) to be "stacked" on top of the compute and SoC components This allows the CPU to This improves the performance of integrated graphics. By combining these tiles, which Intel calls "tiles," chipmakers can create larger chips that can scale the GPU and IO components of a Meteor Lake chip. Intel is also considering ultra-thin 2D materials (three atoms thick) to pack more transistors onto a single chip.

In a press release (opens in new tab), Gary Patton, Intel's GM of Component Research, writes, "75 years after the invention of the transistor, the innovations driving Moore's Law continue to meet exponentially increasing global computing demand." He writes. He went on to say that research advances must "break through current and future barriers" in order to sustain Moore's Law.