Intel Charts Course to Trillion-Transistor Chips: 2D Transistor Materials, 3D Packaging Research

Intel released nine research papers at IEDM 2022 that lay the groundwork for future chip designs as the company looks to deliver on its promise of developing processors with over a trillion transistors by 2030.

Intel’s Components Research (CR) Group lays the initial groundwork for the company’s future technologies, but not all of these initiatives will result in products that ship to market. Those that do come to market would typically arrive in five to ten years.

The group has an incredible track record of innovations that have already come to market, like FinFET, which revolutionized transistor design for the entire industry, strained silicon, Hi-K metal gate, and many others. Intel already has several other technologies on its roadmap, including RibbonFET Gate All Around (GAA) transistors , PowerVia back-side power delivery, EMIB, and Foveros Direct, which all hail from this research group.

The group submitted nine research papers at this year’s 68th-Annual IEEE International Electron Devices Meeting , and below, we’ll cover a few of them in slightly more detail. However, Intel hasn’t yet presented the papers at the conference, so this is broad coverage of the topics.

The pace of transistor density increases continues roughly in line with Moore’s Law, but the economics of today’s chips are not improving at the same pace — the price per transistor is rising as we move to denser nodes. In addition, poor scaling of some chip elements, like analog and caches, complicates matters further. As such, the industry is moving en masse to chiplet-based designs for high-performance chips.

The overriding goal of any chiplet-based design is to preserve the best attributes of the power consumption and performance (latency, bandwidth) of the data pathways inside of a single-die monolithic processor while tapping the economic benefits of using a chiplet-based approach, like increased yield from smaller dies fabbed on a leading-edge process and the ability to use older, cheaper nodes for some of the other functions that see lesser density improvements.

QMC is a new hybrid bonding technique that features sub-3 micron pitches and results in a 10X increase in power efficiency and performance density over the research Intel submitted at last year’s IEDM. That previous paper covered an approach with 10-micron pitches, which was already a 10X improvement. As such, Intel has found a pathway to a 100X improvement in just a few years, showing that the company’s work in hybrid bonding is accelerating rapidly. QMC also enables multiple chiplets to be stacked vertically atop one another, as seen in the graphic above.

Intel’s process roadmap already dips below the nanometer scale to the Angstrom scale, and even though the node naming conventions have long ago lost their relation to actual physical measurements of the transistors, it is clear that a radical new approach will be needed for continued scaling. Most of the industry is betting on a shift to 2D atomic channels in the future , but as with all new tech, there will be many steps to such a radical change.

Today’s chip materials, like silicon, are comprised of three-dimensional crystals, which means atoms are bonded in all three dimensions, thus presenting a fundamental limit to shrinking. In contrast, 2D materials are attractive because all of the atoms are bonded in one plane, thus enabling features to be built with as small as three atoms of thickness.

Enter Intel’s research into 2D materials that it could use for 3D GAA transistors. As a refresher, current GAA designs consist of stacked horizontal silicon nanosheets , with each nanosheet surrounded entirely by a gate. This ‘gate-all-around’ (GAA) technique reduces voltage leakage that prevents switching off the transistors. This is becoming more of an issue as transistors shrink — even when the gate surrounds the channel on three sides, as we see with FinFET transistors.

Intel’s paper describes a Gate All Around (GAA) stacked nanosheet structure with channel materials (nanosheets/nanoribbons) that measure a mere three atoms thick and can operate at room temperature with low leakage current.

Memory in all forms is an integral part of computing, but it also consumes plenty of the power budget at both the chip and system level while also being a limiting factor for performance.

Intel also conducted the world’s first functional demonstration of 3D-stacked ferroelectric memory. The most impressive aspect of this tech is that ferroelectric trench capacitors can be stacked vertically on the logic die atop the transistors. That enables layering the memory atop the logic elements instead of being in its own distinct region, as we see with other types of embedded memory, like SRAM used for L1 and L2 caches.

Ferroelectric memory also enables a similar capability to what we see with NAND flash — the ability to store multiple bits of data in a structure that would typically only store one bit. In this case, Intel demonstrated the ability to store four bits per trench.

In the same vein as the electrical contacts modeling for 2D structures, Intel also shared its modeling efforts for mixed phases and defects for ferroelectric hafnia devices, which will, in turn, further the company’s own research and development processes.

Intel is also researching transistors that ‘don’t forget,’ meaning they don’t lose their data (on/off state) when they lose power. This is akin to any non-volatile storage, like NAND, that can retain its state when power is removed, but it comes in the form of a logic transistor. Intel says it has hurdled two of the three roadblocks to using this technology at room temperature. We’re particularly looking forward to this presentation.

Intel’s other papers at the event outline other research areas, like GaN-on-silicon wafers that can enable future technologies beyond 5G, and better ways to store quantum information to create better qubits for quantum computing.

It’s been 75 years since the transistor altered the course of history, and Intel’s Dr. Ann Kelleher, the VP and GM of Technology Development, will also give a special address at IEDM on Monday. The “Celebrating 75 Years of the Transistor! A Look at the Evolution of Moore's Law Innovation” presentation takes place at 9:45 am PT on Monday, December 5. We’ll follow up with coverage of that presentation soon.