Intel’s 10nm “SuperFin” Makes 18% Leap in Performance—Without Transitioning Nodes

Despite delays in Intel's 7nm process, the semiconductor giant has been improving its 10nm process-based technologies. At Architecture Day 2020, Intel announced a series of breakthroughs, including an updated Willow Cove CPU architecture, Tiger Lake SoC architecture, the discrete Xe-HP GPU, Ice Lake processor, Sapphire Rapids processor, and (perhaps most notably), a new SuperFin transistor technology. 

 

Intel introduces 10nm SuperFin transistor at Architecture Day 2020. Image used courtesy of Intel

 

“We have made great progress with our diverse mix of scalar, vector, matrix, and spatial architectures," explains Raja Koduri Intel's senior VP of architecture, graphics, and software. 

He explains that the new SuperFin—a super FinFET—is "designed with state-of-the-art process technology, fed by disruptive memory hierarchies, integrated into systems with advanced packaging, deployed at hyperscale with lightspeed interconnect links, unified by a single software abstraction, and developed with benchmark defining security features.”

 

A Large Intranode Leap for 10nm

Intel says their latest 10nm transistor is a key enabling technology that is allowing improvements in various new architectures.

Traditionally, significant performance improvements are marked by nodal transition. However, Intel explains that while the new FinFET uses the same 10nm node, it achieves the single largest intranode enhancement in its history—delivering performance improvement (17–18%) compared to a full-node transition.

 

Intel explains that the latest advance in Intel’s transistor technology enables the largest intranode performance improvement. Image used courtesy of Intel

 

SuperFin is an improvement over previous FinFET transistors, in part, because of a redesign using a "SuperMIM capacitor," or a super metal insulator metal capacitor.

SuperFin technology offers enhanced epitaxial growth of crystal structures on source/drain, which increases strain and reduces resistance. Intel claims this allows more current through the channel of the device. In addition, the novel thin barrier reduces via resistance by 30%, enhancing interconnect performance.

The transistor’s gate is improved in two ways. The gate process is improved to drive higher channel mobility, which enhances the frequency response of the device and increases the charging rate of capacitances. Furthermore, this technology offers additional gate pitch options for higher drive current in intensive chip areas.

 

Ultra-thin layers of a new class of high-κ dielectric materials are said to enable a five times increase in MIM capacitance. Image used courtesy of Intel

 

A new class of high-κ dielectric materials is used to increase capacitance by five times within the same footprint compared to industry standards. Thin layers of different high-κ materials, each just a few Angstroms (1 Angstrom = 10-10 meter) thick, are stacked in a repeating “superlattice.” To get an idea of the scale, one silicon atom is approximately 2.2 Angstrom in diameter.

The result is voltage droop reduction and improved product performance. SuperFin transistors offer higher clock speeds at any given voltage and can operate at a lower voltage at any given frequency with a larger dynamic range. 

 

The Goal: Enhance Intel's Chip Performance 

Due to reduced channel resistance and via resistance, this technology significantly reduces the power consumption of chips. Gate improvements ramp up faster processing. Lastly, larger capacitance is said to reduce voltage droop and enable larger driving currents. Each of these performance improvements is targeted to increase the performance of Intel’s chips.   

Leveraging the capability of SuperFin transistors, Xe-HP GPU chips are designed to target data center applications. According to the Intel, Xe-HP is the industry’s first multi-tiled GPU architecture to provide petaflop-scale AI performance and rack-level media performance in a single package based on their EMIB technology.

The company also says Tiger Lake, a "next-gen" mobile processor, and Willow Cove CPU core will also exploit the capabilities of 10nm SuperFin technology.

 

Tiger Lake, Intel’s 11th generation core mobile processors. Image used courtesy of Intel

 

With long delays in the 7nm process, Intel is trying hard to remain competitive with its 10nm process by focusing on transistor improvements. Although Intel’s results look promising, it will be interesting to watch how it competes against AMD and NVIDIA’s offerings in 2021, especially with respect to cost and CPU/GPU capabilities.