Presentation is loading. Please wait.

Presentation is loading. Please wait.

A Survey of Dynamically Reconfigurable FPGA Devices Surbhi Singhal.

Similar presentations


Presentation on theme: "A Survey of Dynamically Reconfigurable FPGA Devices Surbhi Singhal."— Presentation transcript:

1 A Survey of Dynamically Reconfigurable FPGA Devices Surbhi Singhal

2 Basic FPGA structure

3 Selecting a Target FPGA Factors to be considered: Dynamic reconfigurability Reconfiguration time Partial vs Full reconfiguration Granularity

4 Dynamic reconfigurability The ability of a FPGA to modify operation during runtime. Correct FPGA selection very important. The primary advantages of run-time reconfiguration in devices are reduced power consumption, hardware reuse and flexibility. Main problem is the speed of reconfiguration.

5 Granularity Two main architectures: Course grained: smaller number of larger, more powerful logic blocks Advantage: Faster because of easy routing Fine grained : consists of a large number of small logic blacks Advantage: Good utilization Easy conversion to ASIC

6 Xilinx Virtex Architecture Architecture is coarse grained Basic cell of the Virtex FPGA is configurable logic block(CLB) CLB contains circuitry that allows it to efficiently perform arithmetic LUT’s can be configured as SRAM cells Contains programmable input output blocks (IOBs) interconnected to the CLBs

7 Xilinx Configurable Logic Block

8 Configuration of Xilinx Virtex Fully and Partially reconfigurable Module-Based partial reconfiguration  Distinct portions of an FPGA are referred to as reconfigurable modules.  Used for independent design applications  Small-Bit Manipulations  accomplished by making a small change to the design  Switching configuration is fast as bitstream difference is smaller than device difference.

9 Lattice ORCA Architecture Coarse grained architecture Four basic elements: programmable logic cells (PLCs), programmable input/output cells (PIOs), embedded block RAMs (EBRs), and system-level features. Programmable functional unit (PFU) is the basic functional unit containing eight 4-input LUTs, 8 Iatches/FFs and one additional flip-flop for arithmetic functions

10 Lattice OCRA block diagram

11 Configuration of Lattice ORCA Fully and Partially reconfigurable Partial reconfiguration is done by setting a bitstream option that tells the FPGA to not reset the entire configuration Options available to allow one portion of FPGA to remain in operation while the other is being reconfigured Off chip reconfiguration

12 Atmel AT40K Architecture Fine Grained architecture Symmetrical array of identical cells Distributed 10 ns SRAM capability 8-sided core cells with direct horizontal, vertical, and diagonal cell-to-cell connection Small cells lead to large number of cells which leads to greater functionality Each cell can implement 2 Boolean operations of 3 inputs or 1 operation of 4 inputs

13 AT40K Device Overview

14 Configuration of Atmel AT40K Fully and Partially reconfigurable Off chip reconfigeration Configuration data transferred in either Master mode, slave mode or Synchronous RAM mode Master mode:  Auto configuring.  The Master Mode uses an internal oscillator to provide the configuration clock for clocking configuration data Synchronous RAM mode:  Device receives a 32-bit wide bit stream  It is a memory mapped address space.  User has full read/write access

15 Slave mode: configuration is always initiated by an extenal signal.  Slave Serial Mode the device receives serial configuration data  Slave Parallel mode the device receives either 8-bit wide or 16-bit wide parallel data.

16 AItera APEX 20K Architecture Architecture is coarse grained It combines LUT-based, product-term-based and memory into one device Signal interconnections are provided by the Fast Track interconect Consist of an array of MegaLAB structures Each MegaLAB structure consists of a group of logic array blocks (LABs), one Embedded System Block (ESB), and a MegaLAB interconnect Each LE contains a four-input LUT that can implement any function of four variables.

17 APEX 20K interconnect structure

18 MegaLAB structure

19 AItera APEX 20K configuration Only Fully reconfigurable During operation, it stores its data in SRAM cells. Active configuration: both the target and the configuration device generate both control and synchronization signals Passive configuration: device uses microprocessor to control the configuration process Off chip reconfiguration

20 Conclusion Wide variety of dynamically reconfigurable FPGA devices available today Reconfiguration time of partial reconfiguration is much smaller (~4-5 ms) than full reconfiguration(~12 ms)

21 Run-Time FPGA Reconfiguration for Power-/Cost-Optimized Real- Time Systems Marisha Rawlins

22 Introduction Run time partial reconfiguration is useful to designers who want to develop applications demanding adaptive and flexible hardware Using partial reconfiguration can result in power and area savings An intelligent system is needed to manage reconfiguration in order to save power and meet timing constrains in real time systems 22

23 Basic FPGA Layout 23

24 Partial Reconfiguration 24

25 Advantages of Dynamic Reconfiguration Power/Size/Cost Reduction Hardware Reuse Obsolescence Avoidance Application Portability 25

26 Slot Based Dynamically Reconfigurable System 26

27 Basic System Approach Address assignment  Each module is assigned a default address during design time  New modules are assigned valid logic addresses within the legal address range when they are loaded 27

28 Basic System Approach Transferring data from modules  The bus arbiter calls every existing address  If the module’s busy signal is asserted the arbiter selects the next address  The module’s request signal is asserted if the module wants to send data Transferring data to modules  The main controller asserts a Data-In signal if external data is available for the selected module 28

29 Basic System Approach Data transfer schemes  The main controller knows the module’s data transfer time  A time stamp is sent before the data Replacing modules  If a module not currently configured on the FPGA is needed an unused module is replaced  A context-save is done using data and state I/O lines  Data and state is transferred to local memory of the main controller 29

30 Basic System Approach Bus Realization  Interfaces are in a fixed place  All modules can be implemented in every possible function column 30

31 Initial Results Using partial reconfiguration  The application can be implemented on XCV200E instead of XCV300E  Smaller area  Lower power  Lower cost Can meet timing constraints for a real-time system 31

32 Power Consumed During Reconfiguration 32

33 Power Consumed During Reconfiguration 33

34 Conclusions It is possible to use dynamic partial reconfiguration to save both power and area An intelligent run time system is needed to ensure that power is saved and that timing constraints are meet when using partial reconfiguration for real time systems Measuring power consumed during partial reconfiguration aids in determining the design of the run time system and the feasibility of using dynamic partial reconfiguration 34

35 Runtime FPGA Partial Reconfiguration Shaon Yousuf

36 Outline Benefits of Fields-Programmable gate arrays (FPGAs) Overview of Partial Reconfiguration (PR) in FPGAs Partial Reconfiguration in the Xilinx Virtex-4 Software-Defined Radio and Partial Reconfiguration  What is Software-Defined Radio?  Why PR is applicable to Software-Define Radio? PR designs of Software-Defined Radios (SDR) Conclusions

37 Why FPGAs are being used? FPGAs are now being used as replacement for application specific integrated circuits (ASICs) for many space-based applications  FPGAs provide: Reconfigurable Architectures Low Cost solution Rapid Development Times Additional benefits in each new generation other than expected larger size and faster speed

38 Design exploration on Latest FPGAs Exploration and evaluation of designs with latest FPGAs can be difficult  Continuous endeavor as performance and capabilities improve significantly with each release  Each vendors products can have different characteristics and utilities Often there are unique capabilities Thus side by side evaluation of designs among different products are not straightforward One such advancement of particular importance is partial reconfiguration (PR)  Three Vendors provide some degree of this feature Xilinx Atmel Lattice

39 What is Partial Reconfiguration? Partial reconfiguration (PR) allows the ability to reconfigure a portion of an FPGA Real advantage arises when PR is done during runtime also know as dynamic reconfiguration  Dynamic Reconfiguration allows the reconfiguration of a portion of an FPGA while the remainder continues operating without any loss of data  Two types of Regions Static – Keeps operating Reconfigurable – Can be reconfigured with a new module Central Controlling Agent ICAP Mem controller Module A Module B Module C Static modules Reconfigurable Modules (PRMs) FPGA PRR 1 PRR 2 Static region Static modules Modules: A & B Modules: C & D Module D

40 Partial Reconfiguration in Virtex-4 Exploration of Partial Reconfiguration for a design requires significant knowledge on targeted Device Xilinx’s FPGA’s are widespread, so the Virtex-4 Family was chosen as the example FPGA Need to evaluate current performance and limitations  Reconfiguration speeds and methods  Design Hierarchy limitations related to PR Modules (PRMs) and number of allowed PR regions

41 Virtex-4 LX15 FPGA Layout Overall Structure  CLBs – Configurable logic blocks  IOBs – Input-output buffers  DSP48s – Xilinx’s digital signal processing units  BRAMs – Block Random Access Memories  FIFOs – First-in First-out buffers  DCMs – Digital Clock Managers CLBs IOBs DCMs and Clock Dist. DSP48s BRAMS and FIFOs Figure 1. Virtex-4 LX15 FPGA layout

42 Virtex-4 Reconfiguration FPGA is reconfigured by writing bits into configuration Memory (CM) Configuration data is organized into frames that target specific areas of the FPGA through frame addresses To reconfigure any portion of that frame the partial bitstreams contain configuration data for a whole frame Reconfiguration times highly depend on the size and organization of the PR regions  Virtex-II allowed column based PR only  Virtex-4’s allow arbitrarily sized PR regions Virtex-4 Frames  Composed of 41 32-bit words  The LX15 has 3,740 frames Four methods of reconfiguring a device, each has applications where desirable  Externally Serial configuration port JTAG (Boundary Scan) port SelectMap port  Internally Though the Internal configuration access port (ICAP) using an embedded microcontroller or state machine

43 Table 1. Summary of Configuration Options Table 2. Example Configuration Sizes and Times to Configure with JTAG and SelectMAP/ICAP Reconfiguration Speeds Table 1 shows a summary of Configuration Speeds for the four options Table 2 shows example configurations sizes and times for the four options  Values were based on estimates of Xilinx’s PlanAhead Software when targeting an approximate PR region slice utilization of 90%

44 Reconfiguration using an Embedded Microcontroller Many Xilinx FPGAs have Embedded hard processor cores  Processor core has the ability to process C/C++ code  Makes reconfigurable designs extremely flexible since no need for external control Figure 2. PR design using embedded microcontroller Reconfiguration Steps  Reconfiguration is triggered within the FPGA  Processor core loads the desired configuration data from external reconfiguration memory This could be from ROM, Flash, static Ram loaded at startup or filled up by the FPGA itself  Processor reconfigures the PR region through the ICAP primitive

45 PR Design Hierarchy Flow from Hardware Description Language (HDL) to configuration bitstream is extremely complicated  Top-level module should contain sub-modules that are either static or reconfigurable  All communications except global signals such as clock must be explicitly declared using 8-bit bus macros Figure 3. Example design showing two PR regions Current PR design flow allows multiple Partial reconfiguration regions  One Partial Reconfiguration for PRM A and one for PRM B

46 PR Design Hierarchy Cont. Required hierarchy adds significant amount of effort when converting an existing static design into one that is ready for PR Having all PRM at the top level will often require routing many signals to and from another module deep within the main static module Figure 4. Transceiver design with turbo coding and concatenated convolutional + Reed-Solomon coding a) Hierarchical view before PR partitioningb) Required design partitioning

47 PR software Support Initially there was very little software support to assist in generating PR designs and bitstreams Recently  Xilinx’s EA PR flow with the integration of the Xilinx’s PlanAhead tool greatly mitigates the complicacies of PR

48 Software-Defined Radios and Partial Reconfiguration Determining whether PR is appropriate for a given application depends heavily upon the FPGA family One such field PR can be applied for great advantage is software-define radio

49 Why Software Defined Radio? Wireless communication abilities are becoming ubiquitous in the latest generation of portable electronics  Cell phones, Camera’s, MP3 players etc. Satellites need an overwhelming number of communication standards to communicate with these wide array of devices Hardware designs that attempt to provide compatibility with current standards will likely become obsolete shortly after their release Solution: Use Software-defined radios (SDR’s)  SDRs run on a generic hardware platform that allow communication parameters to be defined by the software during runtime  Some common communication functions affected by the parameters are modulation, demodulation, filtering, frequency selection and frequency hopping

50 What is Software-Defined Radio? SDRs split into three basic sections  Radio Frequency (RF) Section Acts as a transceiver, performing conversion up or down to the intermediate frequency  Intermediate Frequency (IF) Section Performs all the necessary signal processing, modulation, demodulation, filtering etc  Base-Band Section Performs the processing of the digital data, i.e, data payload assembled or disassembled Reconfigurable SDR Design  Basically, creating a modular design on an FPGA that can load the desired functions as needed  The reconfigurable nature on an FPGA allows reconfiguring the functionality of a specific block while the reminder of the design continues to function  Provides a unique opportunity to create an extremely flexible and compact design Figure 5. Block Diagram of a generic Software-defined radio

51 PR designs for Software Defined Radios Three SDR types where the application of PR were studied  Simplex Spread-Spectrum Transceiver with FEC  Dynamic Bandwidth Resource Allocation Transceiver  Cognitive Radio

52 Simplex Spread-Spectrum Transceiver with FEC Simplex transceiver  Transmit or receive capabilities used at a given time but never at the same time  Assuming Waveform requires Forward error checking (FEC) Direct-sequence spread spectrum (DSSS)  Two PR regions defined One for Tx modulator or the Rx demodulator One for the Tx FEC encoder, the Rx DSSS acquisition engine, or the Rx FEC decoder Figure 6. Simplex Spread-Spectrum Transceiver with FEC

53 Dynamic Bandwidth Resource Allocation Transceiver Dynamic Bandwidth Resource Allocation (DBRA) systems alter communication waveform dynamically to match channel conditions  Adjust configuration to keep bit error rate at a current threshold Four PR regions Defined  One for Tx Binary Phase Shift-Keying (BPSK) or Tx Gaussian minimum shift keying (GMSK)  One for Turbo encoder or Convolution encoder, followed by Reed-Solomon (RS) encoding  One for Rx BPSK or Rx 8-ary Phase shift-keying (8PSK)  One for Turbo Decoder or concatenated Viterbi, followed by RS decoder Figure 7. Dynamic bandwidth resource allocation transceiver

54 Cognitive Radio Receiver Steps for Cognitive Radio (CR) function  Scan the available spectrum using an FTT  Locate energy  create a channel that attempts to match the spectral shape  Perform modulation recognition  Try to demodulate One PR region  One for Modulation recognition and one for the Demodulator FFT Module is left static as spectrum is frequently monitored  If reconfiguration speed is very fast, FFT can also be made reconfigurable Figure 8. Cognitive radio receiver

55 Conclusions Exploring the usefulness of PR in the field of software define radio has shown both its feasibility and benefits Design time overhead involved when creating a PR design is acceptable but requires progressing though a slow learning curve before any results are obtained Full benefits of PR will not be evident until it becomes commonplace in industry and vendors place more resources on supporting the PR design flow and keeping documentation up to date However, the adaptivity of PR combined with the desire for SDR’s make a strong argument for pursuing PR.

56 Questions?

57 Thanks


Download ppt "A Survey of Dynamically Reconfigurable FPGA Devices Surbhi Singhal."

Similar presentations


Ads by Google