SAN JOSE, Calif. -- 28nm may still be considered the
mainstream node, but for leading-edge designers, there is a clear and
compelling path from there through 16nm and into even the 10nm design
That was the message last week from Suk Lee, Senior Director,
Design Infrastructure Marketing Division, TSMC, who spoke at the annual TSMC
Whether it's design infrastructure (design rules, PDKs, reference
flows, and so forth) or IP, TSMC has charted a course deep into the FinFET era
that will take design teams well into 2020 and beyond.
Some highlights from Lee's presentation:
At this emerging node, DRM, PDKs, libraries, the IP portfolio, and
design-for-manufacturing (DFM) design services from TSMC and its ecosystem
partners is complete. "20nm is there and ready for you," Lee told the
packed ballroom during his presentation.
Among other features, at 16nm, Lee said there is optimized low
Vdd, "which has been a particular focus ... as one of the key enabling
technologies for 16nm," Lee said. In addition, Vt min-area pessimism
reduction has been factored into place and route, he said.
For DRC, TSMC has continued to enhance front- and back-end dummy
fill. For LVS (layout versus schematic), the node features fin layer generation
"to make sure we don't have any spurious problems during LVS time."
In addition, to improve design accuracy and PPA (power,
performance, area), "we've enhanced capabilities to support 3D extraction."
When it comes to electro-migration and IR issues, TSMC has
focused on both gate-level and transistor-level improvements.
For SPICE simulators, TSMC has poly-over-diffusion-edge (PODE) back
annotation support and now has three fast SPICE simulators certified from the
major EDA vendors.
When it comes to a 16FF reference flow for ASIC/SoC design, TSMC
focused on FinFET solutions and methodologies to increase designers' PPA advantage.
"We focused on low-voltage design enablement, EM
reliability, and power integrity management as we go into FinFET-based
design," Lee said. "We also focused on timing analysis and silicon-accurate correlation," using the ARM Cortex-A15 processor as the qualification
vehicle, he added.
For the 16FF custom reference flow, one of the things TSMC took
into account was "rather than continuously variable gate widths, we now
have a number of fins methodology where there are quantized fins available so
we're focused on automatic fin snapping and fin placement, so you can ensure the
design is correctly implemented."
Regarding the 16FF custom integrated tool flow: "We took a
large PLL and ran it through the entire flow in the same manner one of our
customers would. So we've gone through schematic design, simulation, layout,
editing, physical verification, EMI analysis and, finally, RC extraction. 16nm
FinFET full custom design is fully available and ready to go."
Digital integrated tool certification is complete for all major
place-and-route tools at the 16nm node, which emerged a year ago, Lee said.
"We've used ARM Cortex-A15 to qualify all flows. We looked
at single core, quad core designs. We've met our PPA targets using those tools. A lot of work was done by our partners, and we
appreciate the effort."
16 FinFET+ targets a 15% speed gain over 16FF by--among other
ways--boosting DC Ion and Ioff, improving middle end of line capacitance and
back end of line capacitance.
Regarding the design infrastructure supporting 16FF+, TSMC has
expanded its certification program by adding new features to re-validate place-and-route accuracy as well as timing accuracy and EM and IR, Lee said.
He added: "A major task for us as partners has been to
regenerate PDKs and tech files for DRC, LVS, and custom design support. The
reference flows for 16FF+ have digital methodologies in place."
Work on the 10nm node proceeds apace, especially on
Lee said, "The key process challenges for our EDA partners is
the fact that the color A and B are systematically asymmetric on rules and
performance so we have to support full coloring throughout the entire tool
chain and make sure we are balancing proportions of mask A and mask B."
An example of this is the full coloring-aware design solution
for place and route. "In place and route, we've got pre-defined coloring tracks
to match up with pre-defined pins on cells, and this ensures we've got a balance
of mask A and mask B. Furthermore, we have co-optimized cell design and
placement again to improve pin access and to ensure we have mask balancing," Lee
"Since we have asymmetrical behaviors in terms of resistance variation,
we have timing optimization to leverage different wire resistance between masks,"
"We've implemented color-aware design throughout the entire
tool chain," he added.
As for IP cores across the node spectrum, much of the
infrastructure already is in place, with the availability of 6,600 IP titles
from 39 vendors, Lee said. "It continues to grow as we move from process
generation to process generation," he added.
At 28nm, "virtually the entire" IP portfolio is ready
At 20nm, most of the portfolio is ready, with PCIe IP coming in
June, Lee said.
"At 16FF and FF+, the message is 'the portfolio
ready,'" Lee said.
For more information visit TSMC's
--TSMC Forum: 16nm FinFET Design Challenges Met by Custom/Analog Reference Flow
--TSMC OIP Forum: 16nm FinFETs, 3D-ICs Gain EDA and IP Support