From WikiChip
Difference between revisions of "10 nm lithography process"

(10 nm Microarchitectures)
(10 nm Microarchitectures)
Line 210: Line 210:
 
** {{intel|Sapphire Rapids|l=arch}}
 
** {{intel|Sapphire Rapids|l=arch}}
 
** <s>{{intel|Knights Hill|l=arch}}</s>
 
** <s>{{intel|Knights Hill|l=arch}}</s>
 +
* Qualcomm
 +
** {{qualcomm|Falkor|l=arch}}
 
* Samsung
 
* Samsung
 
** {{samsung|Mongoose 2|l=arch}}
 
** {{samsung|Mongoose 2|l=arch}}
 
** {{samsung|Mongoose 3|l=arch}}
 
** {{samsung|Mongoose 3|l=arch}}
* Qualcomm
 
** {{qualcomm|Falkor|l=arch}}
 
 
{{expand list}}
 
{{expand list}}
  

Revision as of 03:49, 5 February 2018

The 10 nanometer (10 nm) lithography process is a semiconductor manufacturing process node serving as shrink from the 14 nm process. The term "10 nm" is simply a commercial name for a generation of a certain size and its technology, as opposed to gate length or half pitch. The 10 nm node is currently being introduced and is set to get replaced by the 7 nm process in 2019.

Industry

At the advanced 10nm process, there are only 3 semiconductor foundries with such manufacturing capabilities: Intel, Samsung, and TSMC.

Due to marketing names, geometries vary greatly between leading manufactures. Although both TSMC and Samsung's 10nm processes are slightly denser than Intel's 14nm in raw logic density, they are far closer to Intel's 14nm than they are to Intel's 10nm (e.g., Samsung's metal pitch just 1 nanometer shorter than Intel's 14nm).


 
Process Name
1st Production
Lithography Lithography
Immersion
Exposure
Wafer Type
Size
Transistor Type
Voltage
 
Fin Pitch
Width
Height
Gate Length (Lg)
Contacted Gate Pitch (CPP)
Minimum Metal Pitch (MMP)
SRAM bitcell High-Perf (HP)
High-Density (HD)
Low-Voltage (LV)
DRAM bitcell eDRAM
Intel TSMC Samsung Common Platform Alliance
The Common Platform Alliance is a joint collaboration between IBM, Samsung, GlobalFoundries, STMicroelectronics, UMC
Paper
P1274 (CPU) / P1275 (SoC) 10FF 10LPE
1st generation; 10 nm Low Power Early
, 10LPP
2nd generation; 10 nm Low Power Plus
, 10LPU
3rd generation; 10 nm Low Power Ultimate
 
2017 June 2017 April 2017  
193 nm 193 nm 193 nm 193 nm
Yes Yes Yes Yes
SAQP SAQP LELELE SADP
Bulk Bulk Bulk Bulk/SOI
300 mm 300 mm 300 mm 300 mm
FinFET FinFET FinFET FinFET
0.70 V 0.70 V 0.75 V 0.75 V
Value 14 nm Δ Value 16 nm Δ Value 14 nm Δ Value 14 nm Δ
34 nm 0.81x 36 nm 0.75x 42 nm 0.88x    
7 nm 0.88x            
53 nm 1.26x            
            20 nm 1.00x;
54 nm 0.77x 66 nm (64 nm*) 0.73x 68 nm 0.87x 64 nm 0.80x
36 nm 0.69x 44 nm (42 nm*) 0.69x 48 nm 0.75x 48 nm 0.75x
0.0441 µm² 0.62x     0.049 µm² 0.61x    
0.0312 µm² 0.62x 0.042 µm² 0.57x 0.040 µm² 0.63x 0.053 µm² 0.65x
0.0367 µm² 0.62x            
               

* - Value reported from IEEE ISSCC/IEDM/VLSI Conference.

Intel

See also: Intel's Process Technology History
intel 10nm fin.png

Announced during Intel's Technology and Manufacturing Day 2017, Intel's 10 nm process (P1274) is Intel's first high-volume manufacturing process to employ Self-Aligned Quad Patterning (SAQP) with production starting in the second half of 2017. Intel detailed Hyper-Scaling, a marketing term for a suite of techniques used to scale a transistor, SAQP, a single dummy gate and contact over active gate (COAG). Intel's initial 10 nm process has up to 60% lower power and 25% better performance than their initial 14 nm but will actually have lower performance than their "14nm++" process. Intel expect their "10nm+" process to surpass that.

Intel's 10nm process is roughly 1.7x the raw logic density of the next densest 10nm process, albeit due to aggressive pattering techniques they also have the most complex process available to date. The process can support multiple threshold voltages, and features 12-metal interconnect layers with the bottom two made of cobalt. This is the first time cobalt is used in a high volume production node. Because of the ever shrinking geometries the wires get smaller each node.

intel interconnect 10 nm.jpg

At 10nm the wires become so small that the barrier layer takes up most of the interconnect, resulting in less space for the copper itself. As the cross section of the wire gets smaller the resistance rises exponentially. Cobalt aims to address this issue, it does not diffuse in the surrounding material, so the barrier layer can be reduced. And even though it has a higher resistance than copper in bulk, it has a two times lower resistance in very small wires. This can be attributed to the larger wires because of the reduced barrier layer and the larger grain size, witch reduces the electron scattering. It also has 10x better resistance to electron-migration.

Intel will leverage their initial 10nm process for their Cannon Lake-based microprocessors which are used exclusively for mobile. They will then utilize their second generation, "10nm+" process, for Ice Lake-based processors which will be used for the mainstream and server platform.

Samsung

ss 14-10nm.png

Samsung demonstrated their 128 Mebibit SRAM wafer from their 10nm FinFET process. Samsung, which unlike Intel uses LELELE (litho-etch-litho-etch-litho-etch), ramped up mass production in May of 2017. ChipWorks/TechInsight measured the CPP/MMP which came a little short of the Common Platform Alliance Paper which was presented in 2016, at 68 nm contacted gate pitch, 51 nm metal pitch, dual-depth shallow trench isolation (STI), and had single dummy gate.

Samsung's initial process was 10LPE (10 Low-Power Early) which was replaced by second generation evolved process 10LPP (10 Low-Power Plus). Samsung intends to introduce a third generational enhanced 10nm process called 8LPP (8 Low Power Plus) which will further improve performance and introduce a small density increase through cell enhancements and a narrower metal pitch. 8LPP improvements over 10LPP is similar to their 11LPP improvements over their 14LPP. It's worth noting that Samsung intends 8LPP to be their last non-EVU node. All subsequent nodes will use EUV.

TSMC

TSMC reported a poly pitch of 64 nm with a metal pitch 42 nm. TechInsight measured them at 66 nm and 44 nm respectively. 10FF is the second process to use FinFET, and is the Industry's first use of Quad-Patterning. This allows for a full node shrink, enabling a 2X increase in logic density compared to their 16nm process. TSMC claims the 10FF process will have 15% higher performance while consuming 35% less power.

10nm tsmc.jpeg

10 nm Microprocessors

This list is incomplete; you can help by expanding it.

10 nm Microarchitectures

This list is incomplete; you can help by expanding it.

Documents

References

  • Mark Bohr, Intel. Intel Technology and Manufacturing Day. Mar 28, 2017.
  • Samsung uses LELELE based on their press release about their 10nm FinFET Technology on October 17, 2016.
  • Seo, K-I., et al. "A 10nm platform technology for low power and high performance application featuring FINFET devices with multi workfunction gate stack on bulk and SOI." VLSI Technology (VLSI-Technology): Digest of Technical Papers, 2014 Symposium on. IEEE, 2014.
  • Cho, H-J., et al. "Si FinFET based 10nm technology with multi Vt gate stack for low power and high performance applications." VLSI Technology, 2016 IEEE Symposium on. IEEE, 2016.
  • Song, Taejoong, et al. "A 10 nm FinFET 128 Mb SRAM With Assist Adjustment System for Power, Performance, and Area Optimization." IEEE Journal of Solid-State Circuits (2016).
  • Clinton, Michael, et al. "12.3 A low-power and high-performance 10nm SRAM architecture for mobile applications." Solid-State Circuits Conference (ISSCC), 2017 IEEE International. IEEE, 2017.
  • Samsung's actual transitor size was measured by ChipWorks/TechInsight based on the Qualcomm Snapdragon 835 which is manufactured on Samsung's 10nm process.
  • TechInsights TSMC 10 nm Process Analysis