LEUVEN (Belgium), December 9, 2019 — This week, at the 2019 IEEE
International Electron Devices Meeting, imec, a world-leading research
and innovation hub in nanoelectronics and digital technologies, presents
first standard cell simulation results of its forksheet device designed
for sub-3nm logic technology nodes. Compared to nanosheet devices, the
reduced n-to-p spacing results in a 10 percent performance increase.
When combined with scaling boosters, the new device architecture will
bring logic standard cell height down to 4.3 tracks, which combined with
cell template optimization can result in more than 20 percent area
reduction. The results value the forksheet architecture as a potential
solution to extend the scalability of nanosheet structures beyond the
3nm logic technology node.
The forksheet device has recently been proposed by imec as a natural
extension of vertically stacked lateral gate-all-around nanosheet
devices. Contrary to the gate-all-around nanosheet device, in the
forksheet, the nanosheets are now controlled by a tri-gate forked
structure, realized by introducing a dielectric wall in between the P-
and NMOS devices before gate patterning. This wall physically isolates
the p-gate trench from the n-gate trench, allowing a much tighter n-to-p
spacing – a challenge that could not be answered with FinFET or
nanosheet structures. Because of this reduced n-to-p separation, the
forksheet is expected to have superior area and performance scalability.
For the first time, standard cell simulations confirm this excellent
power-performance-area (PPA) potential of the forksheet device
architecture. The device under study targets imec’s 2nm technology node,
using a contacted gate pitch of 42nm and a 5T standard cell library with
a metal pitch of 16nm. The proposed design includes scaling boosters
such as buried power rails and wrap around contacts. Compared to a
nanosheet device, a 10 percent speed gain (at constant power) and a 24
percent power reduction (at constant speed) is reported. The performance
boost can be partly explained by a reduced miller capacitance, resulting
from a smaller gate-drain overlap. Finally, the n-to-p separation
reduction can be used to reduce the track height from 5T to 4.3T.
Further layout optimization exploiting the structure of the device
enables more than 20 percent cell area reduction. When implemented in an
SRAM design, the simulations reveal a combined cell area scaling and
performance increase of 30 percent for 8nm p-n spacing.