Intel Corp. has shared new technical particulars about Meteor Lake, an upcoming sequence of desktop and laptop computer processors that can characteristic chiplet expertise.

Intel detailed the processor sequence right now forward of its presentation at Scorching Chips, a significant semiconductor trade occasion.

Intel traditionally carried out all of the transistors that comprise a chip on a single piece of silicon. With the Meteor Lake sequence, the corporate will implement transistors not on a single piece of silicon however reasonably on a number of, smaller modules. This strategy to chip design is understood within the semiconductor trade as a chiplet structure.

Every Meteor Lake processor will characteristic 5 totally different modules, or chiplets. The first chiplet is a central processing unit that Intel will produce utilizing its latest Intel four manufacturing course of, which can be known as the corporate’s seven-nanometer course of. In accordance with the corporate, Intel four makes it attainable to provide CPUs that may function at as much as 21.5% larger frequencies than previous-generation silicon utilizing the identical quantity of electrical energy.

In Meteor Lake processors, the first CPU chipset is built-in with three different chiplets which are made not by Intel however reasonably by Taiwan Semiconductor Manufacturing Co. Ltd. The three chiplets that TSMC will manufacture reportedly embody a graphics processing unit, an I/O module for managing knowledge enter and output operations and a so-called SoC tile. Intel didn’t elaborate on what options the SoC tile gives.

With conventional semiconductor manufacturing strategies, it’s not attainable to mix processing modules made by totally different firms in a single processor. Chiplet expertise facilitates such configurations, which provides chipmakers similar to Intel extra flexibility in how they design their merchandise.

One other good thing about the expertise is that it could actually simplify manufacturing in sure respects. When the chiplets that comprise a processor are produced individually and linked collectively after manufacturing, {hardware} flaws change into simpler to handle. If one of many chiplets comprises a flaw, the others can nonetheless be used.

The CPU, GPU, I/O module and SoC tile in Meteor Lake are all positioned atop one other, fifth chiplet in a three-dimensional configuration. That fifth chiplet reportedly doesn’t include any logic circuits, or circuits able to performing computations. It’s manufactured by Intel utilizing its low-cost 22FFL chip fabrication course of.

Putting chips atop each other in a three-dimensional configuration has a number of main benefits. Two processors positioned atop each other have a smaller space than two processors positioned subsequent two one another, which is helpful for compact gadgets similar to laptops. Furthermore, stacking chips permits knowledge to journey sooner between transistors underneath sure circumstances, which might velocity up processing.

Intel is utilizing an internally developed expertise generally known as Foveros to position Meteor Lake’s 4 most important chiplets on the fifth, 22FFL-based chiplet that serves because the processor sequence’ basis. Intel already carried out Foveros as soon as earlier than in a chip lineup known as Lakefield that it discontinued in 2021. Nonetheless, Meteor Lake is the primary instance of Intel utilizing the expertise to energy a processor sequence that it’ll manufacture in giant volumes.

The primary Meteor Lake chips are anticipated to make their debut subsequent yr.

Along with utilizing Foveros to energy laptop computer and desktop processors, Intel plans to carry the expertise to the info knowledge heart market. The corporate final yr debuted a Foveros-powered chip dubbed Ponte Vecchio that’s optimized for machine studying workloads. The chip contains 47 totally different chiplets made utilizing 5 totally different manufacturing processes.

Throughout its numerous parts, Ponte Vecchio includes a whole of 100 billion transistors. That’s practically double the variety of transistors in Nvidia Corp.’s flagship A100 knowledge heart graphics card. In accordance with Intel, early Ponte Vecchio prototypes have demonstrated the flexibility to achieve speeds of 45 teraflops, which equals 45 trillion computing operations per second.

Picture: Intel