KEYWORDS: Field programmable gate arrays, Logic, System on a chip, Analog electronics, Artificial intelligence, Photomasks, Machine learning, Semiconductors
FPGA (Field Programmable Gate Array) has been continually growing – expanding functionality, improving powerperformance and enlarging capacity, in past several decades. Moore’s Law has been gradually slowing down and technology becomes more expensive in past several nodes. However with continual architecture, design and technology innovations, FPGA’s capacity expansion continues, beyond Moore’s Law. 3D-IC, embedded processing SOC, HBM integration, RF-SOC and furthermore, coming Versal ACAP (Adaptive Compute Acceleration Platform) family, ensures more and more suitable applications in data centers, machine learning (ML), 5G, automotive and many other applications in coming years.
As the semiconductor critical dimension (CD) is shrunk to 20nm node and beyond, double and triple patterning
technologies become necessary for current 193nm optical lithography. However, the new technologies induce a new
variation factor of the two or three mask pattern mismatching in terms of the wafer CD or alignment performance on
silicon. This mismatch can degrade matching circuit performance such as SRAM and analog circuit. In this paper, we
address the impact on our 20nm CRAM (configuration RAM used in FPGA circuit) performance caused by diffusion
layer pattern decomposition (coloring). Furthermore, we propose a methodology to optimize the coloring based on an
alignment performance assessment and CD control of two mask patterns printed on silicon wafer. In the same
experiment, we observed that the OPC (Optical Proximity Correction) is also critical to the coloring methodology. The
silicon results show that after the optimization, the impact of coloring-induced mismatch on CRAM performance can be
reduced significantly.
There are two different foundry-fabless working models in the aspect of mask. Some foundries have in-house mask
facility while others contract with merchant mask vendors. Significant progress has been made in both kinds of
situations. Xilinx as one of the pioneers of fabless semiconductor companies has been continually working very closely
with both merchant mask vendors and mask facilities of foundries in past many years, contributed well in both
technology development and benefited from corporations. Our involvement in manufacturing is driven by the following
three elements: The first element is to understand the new fabrication and mask technologies and then find a suitable
design / layout style to better utilize these new technologies and avoid potential risks. Because Xilinx has always been
involved in early stage of advanced technology nodes, this early understanding and adoption is especially important. The
second element is time to market. Reduction in mask and wafer manufacturing cycle-time can ensure faster time to
market. The third element is quality. Commitment to quality is our highest priority for our customers. We have enough
visibility on any manufacturing issues affecting the device functionality. Good correlation has consistently been
observed between FPGA speed uniformity and the poly mask Critical Dimension (CD) uniformity performance. To
achieve FPGA speed uniformity requirement, the manufacturing process as well as the mask and wafer CD uniformity
has to be monitored.
Xilinx works closely with the wafer foundries and mask suppliers to improve productivity and the yield from initial
development stage of mask making operations. As an example, defect density reduction is one of the biggest challenges
for mask supplier in development stage to meet the yield target satisfying the mask cost and mask turn-around-time
(TAT) requirement. Historically, masks were considered to be defect free but at these advanced process nodes, that
assumption no longer holds true. There is a need to be flexible enough on unrepairable defect at early stage but also a
need for efficient risk management system on mask defect waivers. Mask defects are often waived in low design
criticality area in favor of scrapping the mask and delaying the mask and wafer schedule. Xilinx's involvement in mask
manufacturing has contributed significantly to our success in past many nodes and will continue.
As transistor dimensions become smaller, on-wafer transistor dimension variations, induced by
lithography or etching process, impact more to the transistor parameters than those from the earlier process
technologies such as 90 nm and 130 nm. The on-wafer transistor dimension variations are layout dependent
and are ignored in the standard post layout verification flow where the transistor parameters in a spice
netlist are extracted from drawn transistor dimensions. There are commercial software tools for predicting
the on-wafer transistor dimensions for the improved accuracy of the post-layout verification. These tools
need accurate models for the on-wafer transistor dimension prediction and the models need to be
re-calibrated as the fabrication process is changed. Furthermore, the model-based predictions of the
on-wafer transistor dimensions require extensive computing power which can be time consuming.
In the paper, a procedure to back-annotate the process induced transistor dimension changes into the
post layout extracted netlist using a simple look-up table is described. The lookup table is composed of
specified drawn transistor and its sounding layout as well as their on-wafer dimensions. The on-wafer
dimensions can be extracted from simulations, SEM in-line pictures or electrical data of specially designed
testkeys. Taking the lookup table data, accordingly, the transistor dimensions in the post-layout netlist file
are then modified by a commercial software tool with a pattern search function. Comparing with the
model based approach, the lookup table approach takes much less time for modifying the post-layout netlist.
The lookup table approach is flexible, since the tables can be easily updated to reflect the most recent
process changes from the foundry.
In summary, a lookup table based approach for improving the post-layout verification accuracy is
described. This approach can improve the verification accuracy from both litho and non-litho process
variations. This approach has been applied to Xilinx's 65 nm and 45 nm product developments.
In this paper we describe, from the user's point of view, how Inverse Lithography Technology (ILT) differs from Optical Proximity Correction (OPC). We discuss some specifics of ILT at chip-scale. We show simulation and experimental results from 90nm and 65nm semiconductor nodes, comparing results from ILT-generated masks and OPC-generated masks for real-life layouts, in a production environment. In addition, we discuss issues related to complexity and manufacturability of ILT-generated masks.
In the post-physical verification space called 'Mask Synthesis' a key component of design-for-manufacturing (DFM), double-exposure based, dark-field, alternating PSM (Alt-PSM) is being increasingly applied at the 90nm node in addition with other mature resolution enhancement techniques (RETs) such as optical proximity correction (OPC) and sub-resolution assist features (SRAF). Several high-performance IC manufacturers already use alt-PSM technology in 65nm production. At 90nm having strong control over the lithography process is a critical component in meeting targeted yield goals. However, implementing alt-PSM in production has been challenging due to several factors such as
phase conflict errors, mask manufacturing, and the increased production cost due to the need for two masks in the process. Implementation of Alt-PSM generally requires phase compliance rules and proper phase topology in the layout and this has been successful for the technology node with these rules implemented. However, this may not be true for a mature, production process technology, in this case 90 nm. Especially, in the foundry-fabless business model where the foundry provides a standard set of design rules to its customers for a given process technology, and where not all the foundry customers require Alt-PSM in their tapeout flow. With minimum design changes, design houses usually are motivated by higher product performance for the existing designs. What follows is an in-depth review of the motivation to apply alt-PSM on a production FPGA, the DFM challenges to each partner faced, its effect on the tapeout flow, and how design, manufacturing, and EDA teams worked together to resolve phase conflicts, tapeout the chip, and finally verify the silicon results in production.
In this paper we describe, from the user's point of view, how Inverse Lithography Technology (ILT) differs from Optical Proximity Correction (OPC). We show simulation and experimental results from 90nm and 65nm semiconductor nodes, comparing ILT-generated masks and OPC-generated masks for real-life layouts, in a production environment. In addition, we discuss issues related to complexity and manufacturability of ILT-generated masks.
At the sub 90nm nodes, resolution enhancement techniques (RETs) such as optical proximity correction (OPC), phase-shifting masks (PSM), sub-resolution assist features (SRAF) have become essential steps in the post-physical verification 'Mask Synthesis' process and a key component of design for manufacturing (DFM). Several studies have been conducted and the results have been published on the implication and application of the different types of RETs on mask printability and costs. More specifically, double-exposure-based, dark-field, alternating PSM (Alt-PSM) technology has received lot of attention with respect to the mask manufacturing challenges and its implementation into a production flow, despite its yield and critical dimension (CD) control superiority.
Implementation of Alt-PSM generally requires phase compliance rules and proper phase topology in the layout and this has been successful for the technology node with these rules implemented. However, this may not be true for a matured, production process technology, in this case 90 nm. Especially, in the foundry-fabless business model where the foundry provides a standard set of design rules to its customers for a given process technology, and where not all the foundry customers require Alt-PSM in their tapeout flow. What follows is an in-depth review of the DFM challenges to each partner faced, its effect on the tapeout flow, and how design, manufacturing, and EDA teams worked together to resolve phase conflicts, tapeout the chip, and finally verify the silicon results in production.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.