Reconfigurable FPGAs The Xilinx Virtex II Pro ProX FPGA family

1
Reconfigurable FPGAs (The Xilinx Virtex II Pro /
ProX FPGA family)

• Shadab Ambat

03/02/2006
2
Introduction

• Virtex-II Pro FPGAs were incorporated in 2002 and
  fabricated in 130 nm, 1.5V process technology.
• They incorporate upto 2 embedded IBM PowerPC (400
  MHz) processors and 3.125 Gbps RocketIO serial
  transceivers.
• The, Virtex-II Pro X is an extended version,
  introduced in 2003 with a transceiver data rate
  to 6.25 Gbps.

3

• They can be scaled down or up in features,
  density, I/O and performance depending on the
  requirements.
• Embedded and distributed memory
• Digital Clock Management for on-chip/off-chip
  clock synthesis and synchronization
• XCITE digitally-controlled I/O impedance to
  improve signal integrity and reduce board space
• Full/partial FPGA reconfiguration
• Up to 444 18X18 bit embedded multipliers
• Extensive library of DSP algorithms
• DSP tools such as The MathWorks MATLAB/Simulink,
  the Xilinx System Generator for DSP, and Cadence
  SPW
• Over 200 IP cores from Xilinx and partners

4
Virtex II ProX

• Two devices - 2VPX20 and 2VPX70
• Has all features of the Pro version
• Additional features include new embedded RocketIO
  X serial transceivers consisting of-
• Eight or 20 channels per device
• 2.488 to 6.25 Gbps per channel
• 20X and 16X clocks and data paths
• 8b/10b and 64b/66b encoding
• SONET/SDH OC-48 jitter compliance
• Programmable receiver equalization for enhanced
  signal integrity
• Ideally suited for 10G Ethernet, SONET OC-48, and
  proprietary backplane applications

5
Architecture Overview

• Apart from the RocketIO cores and the RISC CPU,
  the FPGA consists of the following
• Input/Output Blocks (IOBs) They are
  bidirectional blocks that can be programmed as
  inputs or outputs. Inputs and outputs can have
  optional SDR or DDR data registers and outputs
  can also have optional tri-state buffers.
• Configurable Logic Blocks (CLBs) FPGAs consist
  of an array of CLBs arranged in rows and columns.
  These are the blocks that implement the
  sequential and combinatorial logic of the FPGAs

6

• Block SelectRAM Memory They consist of 18 Kb of
  true Dual-Port RAM, programmable from 16K x 1 bit
  to 512 x 36 bit, in various depth and width
  configurations.
• 18 X 18 Bit Multipliers A multiplier block is
  associated with each SelectRAM memory block.
• The multiplier block is a dedicated 18 x 18-bit
  2s complement signed multiplier.
• Read/multiply/accumulate operations and DSP
  filter structures are extremely efficient.
• Global Clocking Implemented using Digital Clock
  Manager (DCM) and global clock multiplexer
  buffers.
• Upto 12 DCMs available, providing deskewed and
  90-, 180- and 270- degree phase shifted versions
  of its output clocks.

7

• Routing Resources All the above blocks use the
  same interconnect scheme and the same access to
  the global routing matrix.
• Boundary Scan Boundary-scan (JTAG) instructions
  and associated data registers support a standard
  methodology for accessing and configuring
  Virtex-II Pro devices.
• Can function in system mode, in which the device
  will continue to function while executing
  non-test boundary-
• scan instructions or in test mode in which
  boundary-scan test instructions control the I/O
  pins for testing purposes.
• Configuration Virtex-II Pro / Virtex-II Pro
  devices are configured by loading the bitstream
  into internal configuration memory using one of
  the following modes
• Slave-serial mode
• Master-serial mode
• Slave SelectMAP mode
• Master SelectMAP mode
• Boundary-Scan mode (IEEE 1532)

8

• Readback and Integrated Logic Analyzer
  Configuration data stored in device configuration
  memory can be read back for verification (using
  Xilinx ChipScope Integrated Logic Analyzer cores
  and Integrated Bus Analyzer cores with the
  corresponding software).

9
Configurable Logic Blocks

• A conceptual model of the FPGA is shown in the
  fig. with its 3 basic elements namely the IOBs,
  Programmable Interconnects and most importantly
  the CLBs.
• Although FPGAs depend more heavily on
  interconnects than CPLDs, the logic functions are
  mainly realized by the CLBs.

10

• CLB resources consist of four slices and two
  3-state buffers.
• Each slice is equivalent and contains
• Two function generators (F G)
• Two storage elements
• Arithmetic logic gates
• Large multiplexers
• Wide function capability
• Fast carry look-ahead chain
• Horizontal cascade chain (OR gate)
• The function generators F G are configurable as
  4-input look-up tables (LUTs), as 16-bit shift
  registers, or as 16-bit distributed SelectRAM
  memory.

11

• The two storage elements are either
  edge-triggered D-type flip-flops or
  level-sensitive latches.
• Each CLB has internal fast interconnect and
  connects to a switch matrix to access general
  routing resources.
• Each slice has a dedicated OR gate for
  implementing a Sum of Products chain.

12
Fig. A Virtex II Pro Slice (Top Half)
13
IP Cores

• The Virtex II Pro FPGA supports several hardware
  and software Intellectual Property cores. Some of
  the prominent ones are
• Hardware Cores
• Bus Infrastructure cores (arbiters, bridges etc.)
• Memory cores (like DDR, Flash)
• Peripheral cores (UART, IIC)
• Networking cores (e.g. ATM, Ethernet)
• Software Cores
• Boot code
• Test code
• Device drivers
• Protocol stacks
• RTOS integration
• Customized board support package

14

• The IP cores are modular, portable, RTOS
  independent, and CoreConnect compatible hence
  allowing design migration.

15
18bit 18 bit Multipliers

• The device has several 18-bit by 18-bit 2s
  complement signed, dedicated embedded multiplier
  blocks.
• The multipliers can be associated with an 18 Kb
  block SelectRAM resource or can be used
  independently.
• They are optimized for high-speed operations and
  have a lower power consumption compared to the
  18-bit x 18-bit multiplier in the CLB slices.
• Each memory and multiplier block is tied to four
  switch matrices
• The SelectRAM memory can be used only up to 18
  bits wide when the multiplier is used, because
  the multiplier shares inputs with the upper data
  bits of its memory.

16

• Both A and B are 18-bit-wide inputs, and the
  output is 36 bits.

Fig. The Memory and multiplier block connections
Fig. A multiplier block
17
Device Configuration

• The FPGA is configured by loading application
  specific configuration data into the internal
  configuration memory.
• Configuration is carried out using a subset of
  the total pins, some of which are dedicated,
  while others can be re-used as general purpose
  I/Os once configuration is complete.
• There are 3 mode pins M0, M1 and M2 which select
  out of five configuration modes.
• The mode pins should be either pulled-up or down
  through resistors, or tied directly to ground or
  VCCAUX (2.5V) to prevent varying supply during
  configuration, and should not be toggled during
  and after configuration.

18
Configuration Modes

• Virtex-II Pro has five configuration modes
• Slave-Serial Mode
• Master-Serial Mode
• Slave SelectMAP Mode
• Master SelectMAP Mode
• Boundary-Scan (JTAG, IEEE 1532) Mode

19

• Slave-Serial Mode
• In slave-serial mode, the FPGA receives
  configuration data in bit-serial form from a
  serial PROM or other serial source of
  configuration data.
• The CCLK pin on the FPGA is an input in this
  mode.
• CCLK is externally generated
• Multiple FPGAs can be daisy-chained for
  configuration from a single source.
• Master-Serial Mode
• In this mode, it is the Virtex-II Pro FPGA device
  that drives the configuration clock (CCLK)
• The interface is identical to slave serial except
  that an internal
• oscillator is used to generate the configuration
  clock.
• A wide range of frequencies can be selected for
  CCLK which always starts at a slow default
  frequency.

20

• Slave SelectMAP Mode
• This is the fastest configuration option.
• Byte-wide data is written into the device with a
  BUSY flag controlling the flow of data. An
  external data source provides a byte stream,
  CCLK, an active Low
• Chip Select (CS_B) signal and a Write signal
  (RDWR_B). If BUSY is asserted (High) by the FPGA,
  the data must be held until it goes Low.
• Data can also be read using the SelectMAP mode.
  If RDWR_B is asserted, configuration data is read
  out of the FPGA as part of a readback operation.
• Master SelectMAP Mode
• This is a master version of the previous mode.
• The device is configured byte-wide on a CCLK
  supplied internally by the FPGA. Timing is
  similar to the Slave SerialMAP mode except that
  CCLK is supplied by FPGA.

21

• Boundary-Scan (JTAG, IEEE 1532) Mode
• In boundary-scan mode, dedicated pins are used
  for configuring the device. The configuration is
  done entirely through the IEEE 1149.1 Test Access
  Port (TAP).
• Configuration through the boundary-scan port is
  always available, independent of the mode
  selection.

22
Readback

• In this mode, configuration data from the
  Virtex-II Pro FPGA device can be read back.
• Readback is supported only in theSelectMAP
  (master and slave) and Boundary Scan mode.
• Along with the configuration data, it is possible
  to read back the contents of all registers,
  distributed SelectRAM, and block RAM resources
  and is useful for debugging purposes.

23
Partial Reconfiguration

• Partial reconfiguration can be accomplished in
  either Slave SelectMAP mode or Boundary-Scan
  mode.
• New data is loaded into a specified area of the
  chip, while the rest of the chip remains in
  operation.
• Chip is not required to be reset as in a full
  configuration
• Data is loaded on a column basis, with the
  smallest load unit being a configuration frame
  of the bitstream (device size dependent).
• Partial reconfiguration is useful for
  applications that require different designs to be
  loaded into the same area of a chip, or that
  require the ability to change portions of a
  design without having to reset or reconfigure the
  entire chip (dynamic reconfiguration).