High speed pattern streaming system based on AXIe’s PCIe connectivity and synchronization mechanism
By Hank Lin, Product Manager of ADLINK Technology, Inc.
E-Beam (Electron Beam) lithography is a next-generation maskless semiconductor manufacturing process. Being maskless, it eliminates diffraction limitations found in current photolithography for 20nm or smaller processes. E-Beam lithography requires a data streaming system with very high throughput that simultaneously transfers massive amounts of pattern data from a data server, processes the data, while outputting to an E-Beam tool. The E-Beam tool itself interfaces through thousands of fiber optic channels and requires channel-to-channel skew less than 2ns. Based on the challenging density and performance requirements, ADLINK chose the
AdvancedTCA Extensions for Instrumentation and Test (AXIe) architecture to implement E-beam streaming. This article describes how the AXIe platform was leveraged to achieve these demanding density and synchronization requirements.
As noted above, E-Beam lithography eliminates diffraction limitations found in traditional photolithography. The concept is much like a printer, but at
breathtaking speed. Instead of ink, thousands of parallel electron beamlets write a pattern directly onto surface of the wafer, itself covered with photoresist. The E-Beam system is controlled through 8,000 optical channels that control a MEMS blanker array, essentially turning an individual beam on or off. The resolution of the lithography demands that the channel to channel skew between the 8,000 optical channels not exceed 2ns.
Manufacturing throughput demands 10 wafers to be processed per hour, or one every 6 minutes. Each mask file can have as much as 2.5TB, which must be transferred to the pattern streaming system in real time. This data is then
processed by the system, which is turn drives the 8,000 optical channels controlling
Why AXIe?
AXIe is an open standard that extends AdvancedTCA* (Advanced Telecom
Computing Architecture) for high performance instrumentation. Particular aspects of AXIe that enabled ADLINK to meet these stringent requirements were:
• A large board size that provides sufficient space for high-density fiber optical channels.
• Designed for high-power applications of up to 200 Watts per slot, utilizing a single rail power supply.
• Good thermal performance with high performance air cooling system
• High-speed PCIe (PCI Express) data fabric
• Flexible scalability and rack-space efficiency. One AXIe chassis can contain one to 14 slots arranged vertically or horizontally, and multiple chassis can be
configured to implement a high-channel-count synchronized system.
• Hardware platform management features, including Shelf Management
Controller, Intelligent Platform Management Controller, and hot swap capability
• Synchronization and local bus features bring precision clocks to each slot
Figure 1: AXIe Bus Distribution
Pattern Streaming Structure
The Pattern streaming system comprises a computer module, PCIe switch module, multiple streamer modules, a 14-slot AXIe chassis, an external synchronizer, and a RAID box as shown in Figure 2.
The computer module accesses the data file from the data center, and then loads the RAID box with the patterns for a specific wafer through the 6Gb SAS link, which it then accesses during the actual lithography. The switch module is a PCIe fabric switch placed in the hub slot. It routes the high-speed data coming from the computer module to the appropriate streamer module. Each of the 12 streamer modules can support 72 optical fiber output channels. The synchronizer is used to synchronize multiple chassis by using clock and trigger distribution.
Most of the unique features of AXIe are utilized in the pattern streamer system, including PCB outline and assemblies, hardware platform management and
monitoring mechanism, power distribution scheme, active cooling system, and data transfer interface.
Horizontal and vertical alignment used in the E-Beam system requires more
complicated synchronization, and deploys the AXIe STRIG and SYNC signals. This point to point trigger system allows very precise and low jitter synchronization to each slot.
Channel-to-channel skew
For the E-Beam system, a maximum 2ns channel-to-channel skew is guaranteed by
the hardware design. Starting at the synchronizer, low skew clock fan-out buffers are
used to distribute the clock and synchronization signals from the synchronizer to the
switch module in each chassis. Besides being the hub point for the PCIe, the switch
module is also the hub for the STRIG, SYNC, and associated clock signals. These
distribute the timing signals to each slot where the streamer modules reside. On the
streamer modules, care is taken to use equal trace lengths for all timing and data
signals while utilizing low-skew buffers. Avago parallel-fiber-optic transmitters
(AFBR-810BHZ-TX) are deployed for the final optical generation. After considering
the skew effect of FPGA, optical fiber, connector and PCB trace, calculated maximum
channel-to-channel skew is 1ns.
Figure 2: Streaming system architecture
Figure 3: Single AXIe chassis fully loaded with 12 streamer modules
Data Center
...
10G Ethernet
Computer Blade
Switch Board
AXIe
Streamer AXIe
Streamer
14-slot AXIe Chassis 12 Pattern Streamers
14-Slot AXIe Chassis
...
10x Chassis
...
System Capacity:
10 Chassis, 10 RAID, 10G Ethernet to Data Center 120 pattern streamers 360 Stratix IV FPGAs 8640 optical transmitters
RAID
SAS/PCIe
RAID
SAS/PCIe
Clock & Trigger Distribution
Amplifier Trigger
Reference Clock
...
PCIe
PCIe PCIe
...
Computer Blade
Switch Board
AXIe
Streamer AXIe
Streamer
12 Pattern Streamers PCIe
PCIe PCIe
Figure 4: System block diagram
High throughput of Pattern Streaming
Besides the tight timing required across 10 chassis, the system also demands massive real time data flows to the optical channels. Four powerful FPGAs on each pattern streamer module include one at the PCIe interface, and three that drive 24 optical channels each, or 72 within a single slot.
The pattern is first read out from the RAID storage to the computer blade memory.
It is then transferred via PCIe DMA (Direct Memory Access) to the appropriate
streamer module. The PCIe FPGA on the streamer module performs the receiving
DMA and stores the data onto onboard flash buffers. It is then transferred to DDR3
DIMM storage associated with each pattern streaming FPGA. The pattern streaming
Optical TX
AFBR-810BPZ
PCIe Gen1 x4 8GB DDR3
ECC SDRAM 8GB DDR3
ECC SDRAM
FPGA4 PCIe IF
EP2AGX65DF29C5N
FPGA2
EP4SGX360KF40C2N
GHz LVDS
16GBx6 Flash
Buffer Buffer AXIe Clock AXIe Trigger Bus
8GB DDR3 ECC SDRAM 8GB DDR3
ECC SDRAM
8GB DDR3 ECC SDRAM 8GB DDR3
ECC SDRAM
72bits@400MHz 72bits@400MHz
72bits@400MHz 72bits@400MHz
72bits@400MHz 72bits@400MHz
4 Gbpsx12
4 Gbpsx12
4 Gbpsx12
4 Gbpsx12
4 Gbpsx12
4 Gbpsx12
GHz LVDS
GHz LVDS
PATA
FPGA3
EP4SGX360KF40C2N
FPGA1
EP4SGX360KF40C2N
Optical TX
AFBR-810BPZ
Optical TX
AFBR-810BPZ
Optical TX
AFBR-810BPZ
Optical TX
AFBR-810BPZ
Optical TX
AFBR-810BPZ
Intelligent Platform Management
Controller
IPMB
Memory Buffer
Figure 5: Pattern Streamer Board Block Diagram
The DDR3 memory is portioned into two banks, which allows a “ping-pong”
technique to optimize the read/write bandwidth. Each optical channel pattern size
can be up to 300MB, or 260GB for a fully loaded chassis. Completion of the file
transfer in 6 minutes requires an aggregate bandwidth of 725MB/s, which is
supported through combined architecture of the computer module, PCIe switch
fabric, and pattern streaming modules. It should be noted that a pattern for a completely different wafer may be downloaded while the current pattern is driving the E-beam machine. This allows for high throughput manufacturing in a high mix environment.
Summary
ADLINK’s FPGA-based AXIe pattern streaming system provides an extremely effective solution for pattern data processing and storage for maskless E-Beam applications, delivering less than 2ns overall channel-to-channel skew and production capable data transfer throughput. The AXIe system architecture was critical in delivering both high aggregate data rates along with tight timing synchronization across multiple chassis. AXIe also delivers the power, cooling and board area needed for the high speed electronics on each streaming module along with high reliability and scalability.
*AdvancedTCA, Advanced Telecom Computing Architecture, and PICMG are all trademarks of the PCI Industrial Manufacturers Group
Hank Lin is Data Acquisition Product Manager for ADLINK Technology’s Measurement andAutomation Products Segment. Hank received his Master’s in Engineering Science and Ocean Engineering from the prestigious National Taiwan University in 1999, before joining ADLINK’s Application Engineering team and then transferring to the Test And Measurement Product Center as Product Manager, developing our Pattern Streaming system shortly thereafter.
Contact Info:
ADLINK Technology, Inc./ AXIe Consortium [email protected]
www.adlinktech.com /www.axiestandard.org
About ADLINK
ADLINK Technology provides a wide range of embedded computing products and services to the test & measurement, automation & process control, gaming,
communications, medical, network security, and transportation industries. ADLINK products include PCI Express-based data acquisition and I/O; vision and motion control; and AdvancedTCA, CompactPCI, and computer-on-modules (COMs) for industrial computing. With the acquisition of Ampro Computers, Inc. and LiPPERT Embedded Computers GmbH, ADLINK also provides a wide range of rugged by design Extreme Rugged™ and Rugged product lines including single board computers, COMs and systems.
ADLINK strives to minimize the total cost of ownership (TCO) of its customers by providing customization and system integration services, maintaining low
manufacturing costs, and extending the lifecycle of its products. ADLINK is a global company with headquarters and manufacturing in Taiwan; R&D and integration in Taiwan, China, the US, and Germany; and an extensive network of worldwide sales and support offices.
ADLINK is ISO-9001, ISO-14001, ISO-13485 and TL9000 certified, is an Associate Member of the Intel® Intelligent System Alliance, an Executive Member of PICMG, a Sponsor Member of the PXI Systems Alliance, an Executive Member of PC/104 Consortium, and a Strategic Member of the AXIe Consortium. ADLINK is a publicly traded company listed on the TAIEX Taiwan Stock Exchange (stock code: 6166).
