Interconnect Testing in Cluster Based FPGA Architectures - PowerPoint PPT Presentation

About This Presentation
Title:

Interconnect Testing in Cluster Based FPGA Architectures

Description:

... in Cluster Based FPGA Architectures. Research by Ian G.Harris and Russel Tessier ... Cluster-based architectures become an architecture of choice for a lot ... – PowerPoint PPT presentation

Number of Views:101
Avg rating:3.0/5.0
Slides: 19
Provided by: moniqueu
Category:

less

Transcript and Presenter's Notes

Title: Interconnect Testing in Cluster Based FPGA Architectures


1
Interconnect Testing in Cluster Based FPGA
Architectures
  • Research by Ian G.Harris and Russel Tessier
  • University of Massachusetts.
  • Presented by
  • Alpha Oumar Barry
  • McGill University MACS
  • 304-649B
  • March 2001

2
Motivation
  • IC densities increase
  • More complex FPGA Architectures
  • Cluster-based architectures become an
    architecture of choice for a lot of FPGA
    manufacturers.

3
Definition / Challenge
  • Several logic blocks are grouped together into a
    cluster. Clusters are connected to other clusters
    via Switch Matrices.
  • High density local interconnect within clusters
  • - improves FPGA utilization
  • BUT!!! - greatly complicates FPGA testing
  • problem.

4
FPGA composed of an array of identical tiles
(Fig. 1a).
Tile composed of a cluster and surrounding
interconnect (Fig 1b).
Interconnect set of lines which can be connected
by a set of programmable interconnect points
(PIP) which act as switches. Fig 1c is a
commonly used structure for PIPs. Each dashed
line is a PIP. A PIP connects each line to one
line on each side of the matrix. For testing
purposes we need to make a difference between
cluster I/O and tile I/O. Cluster I/O are
internal to the tile while tile I/O connects to
neighboring tiles.
5
Cluster composed of a set of basic logic
elements (BLE)
BLE composed of a set of programmable lookup
tables (LUT), multiplexers and flip- flops.
Each BLE input can connect to the output of any
other BLE and to any cluster input. The output
of each BLE is assumed to be connected directly
to a cluster output.
6
Testing Methodology
  • Hierarchical approach defines several FPGA
    configurations that detect bridging faults
    involving intra-cluster and extra-cluster
    interconnects
  • Hierarchical structure of a cluster-based tile is
    exploited in defining the different
    configurations (To facilitate test configuration
    generation process).

7
Testing Methodology
  • In each configuration, FPGA circuitry dedicated
    as BIST logic will perform test generation and
    response analysis to test non-BIST FPGA circuitry
  • Consequently, FPGA will be configured as many
    independent BISTers structure similar to the one
    in Fig. 3

8
BISTer composed of a test pattern generator
(TPG), an output response analyzer (ORA), and two
blocks under test (BUT).
TPG simply a counter which applies an exhaustive
test sequence to the BUT BUT each BUT is a
single tile in the FPGA which is being
tested ORA a comparator which sets the
Pass/Fail flip-flop to 1 if the outputs of both
BUTs do not agree (XOR). Each BISTer will be
implemented as a rectangular block of tiles. The
number of tiles in a BISTer will depend on the
number of tiles needed to implement TPG and ORA.
BISTers will be implemented to cover an entire
tile array.
9
!!! Tiles dedicated to TPG and ORA !!! Are not
completely tested. gtgt FPGA will be reconfigured
to shift the BISTers across the entire
array (Fig. 4)
Tile is fully tested when it acts as a BUT. But
BUT must be surrounded by TPG/ORA. !!!
Perimeter tiles will not be fully tested (by
this uniform repetitive reconfiguration
procedure !!! gtgt BISTer layout must be modified
to use I/O pads to access the tiles on the
periphery
10
Interconnect Faults
  • Permanent Connection (PC) Short between two
    lines
  • Permanent Disconnection (PD) Line broken in two
    or more lines
  • Stuck-At 0 (SA0) PC of a line and ground
  • Stuck-At 1 (SA1) PC of a line and power

11
Observability/Controllability
  • PC defect Both lines separately controllable
  • At least one line observable
  • PIP configured off
  • PD defect Both lines controllable
  • Both lines observable
  • PIP configured on
  • SA0/SA1 Line must be controllableobservable

12
Intra-Cluster Configuration
  • LUT configured as 4-input XOR
  • All BLE outputs are separately controllable from
    each other, and from all cluster inputs
  • Each IMUX configured to select data from each of
    its inputs in at least one configuration
  • There is a sensitized path from each cluster
    input stem to a cluster output in each conf.

13
Algorithm 1 Intra-cluster Configuration Algorithm
  • label all intra-cluster faults as undetected
  • repeat
  • repeat
  • select a BLE which is not configured, b
  • initialize IMUX configurations of b
  • repeat
  • enumerate next IMUX configuration
  • compute BLE output function
  • until BLE function is unique
  • until all BLEs are configured
  • identify detectable faults
  • until all faults are detectable in some
    configuration

14
Extra-Cluster Configurations
  • Goal create current flow paths between tile I/O
    nodes which allow the detection criteria for each
    fault to be satisfied in at least one
    configuration

15
Algorithm 2 Extra-cluster Configuration Algorithm
  • Create interconnect graph
  • Repeat
  • Label all nodes as untouched
  • Repeat
  • Select an untouched node n
  • Identify an untouched path from n to a
    cluster I/O
  • label all nodes on the path as touched
  • Identify an untouched path from n to a
    tile I/O
  • label all nodes on the path as touched
  • Until paths connect all cluster I/O to tile
    I/O
  • Repeat
  • Select an untouched node n
  • Identify an untouched path from n to a
    tile I/O
  • label all nodes on the path as touched
  • Identify an untouched path from n to a
    tile I/O
  • label all nodes on the path as touched
  • Until no more untouched paths can be created
  • Identify detectable faults
  • Until all faults are detectable in a configuration

16
Experimental Results
N Number of BLEs Confs Number of FPGA
configurations I Number of Cluster Inputs Min
Theoretical minimum number of
configurations Fcov Fault coverage (One
additional configuration would give 100 fault
coverage in all cases)
17
Advantages
  • This BIST strategy decomposes testing problem of
    entire FPGA into many identical problems of a
    size which is fixed by the test requirements for
    a single tile (design of TPG and ORA is same for
    all tiles!)
  • Size of smaller problem is fixed, so this
    approach is easily scalable to FPGA arrays of any
    size

18
Advantages / Disadvantages
  • Many ASIC DFT approaches involve modifying the
    circuit functionality to incorporate test
    functionality
  • FPGA reconfiguration avoids that circuit overhead
  • BUT!!! Several configurations are needed and time
    to reconfigure FPGA is not negligible
Write a Comment
User Comments (0)
About PowerShow.com