Title: CSET 4650 Field Programmable Logic Devices
1CSET 4650 Field Programmable Logic Devices
Antifuse-Based FPGAs Actel
2Antifuse FPGAs
- One-time programmable devices
- Primary vendors
- Actel
- QuickLogic
- No longer producing antifuse devices
- Xilinx
- Cypress
- Focus on Actel today
3New Architecture
- Actel FPGAs have evolved from the channeled logic
array architecture (e.g., used in ACT series) to
the Sea-of-Modules architecture - Made possible because antifuses and wires can be
fabricated above the chip floor in the third
dimension
4Antifuse Switch
- Antifuses are originally open circuits that take
on low resistance only when programmed. - Antifuses are manufactured using modified CMOS
technology.
5Antifuse Switch
- Here is a three-dimensional representation of the
same type of routing resources - Some contacts shown are permanent, not switched
- Programmable antifuses are shown in green
- Routing elements of this type allow more density
of logic elements in silicon
6ONO Antifuse Actel
- The figure below illustrates Actels PLICE
(programmable logic interconnect circuit element)
antifuse structure. - The antifuse, positioned between two interconnect
wires, consists of three sandwiched layers - conductors at top and bottom
- an insulator in the middle
7ONO Antifuse Actel
- Unprogrammed, the insulator isolates the top and
bottom layers programmed, the insulator becomes
a low-resistance link. - PLICE uses polysilicon and n diffusion as
conductors and a custom-developed compound, ONO
(oxide-nitride-oxide), as an insulator. - Other antifuses rely on metal for conductors,
with amorphous silicon as the middle layer.
8ONO Antifuse Actel
- An antifuse is the opposite of a fuse.
- It is an open circuit until a current is forced
through it (about 5 mA). - The current melts a thin insulating layer to form
a thin permanent and resistive link.
(a) A cross section. (b) A simplified drawing.
The ONO (oxidenitrideoxide) dielectric is less
than 10 nm thick, so this diagram is not to
scale. (c) From above, an antifuse is
approximately the same size as a contact.
9Antifuse Advantages
- Highest density
- a simple cross point
- 10X the density of SRAM
- Lowest switch resistance
- 25 Ohms
- Very low capacitance
- 1 fF per node
- approaching the metal line capacitance
- non-volatile
- Nearly impossible to reverse engineer
- Radiation hard
- Live within 1 millisecond of the power supply
reaching spec voltage - Software is easy to place and route
10Antifuse Disadvantages
- ONO antifuses require only about 5mA for
programming - 10mA for Metal antifuses
- Some antifuse defects not testable until
programming - hence only 98 to 99 programming yield
- Requires programmer
- Requires a socket
- a problem for devices with more than 200 pins
- Those who design by test will throw out a lot of
parts - Requires one or two transistors per wire for
programming
11FPGA Qualitative Comparison
12Actels Current Antifuse Devices
- Axcelerator
- High-speed antifuse FPGAs with gate densities of
up to 2 million equivalent gates - SX-A / SX
- Antifuse devices 8k to 72k gates
- eX
- Antifuse devices 3k to 12k gates
- MX
- Antifuse devices 3k to 54k gates
13Actels Legacy Devices
- Integrator Series FPGAs
- 1200XL and 3200DX Families v3.0
- Accelerator Series FPGAs
- ACT 3 Family (-3 speed grade only)
- ACT 2 Family FPGAs v4.0.1
- ACT 1 Series FPGAs
Devices being phased out or already discontinued.
Speed Grade Std Standard Speed 1
Approximately 15 faster than Standard 2
Approximately 25 faster than Standard 3
Approximately 35 faster than Standard P
Approximately 30 faster than Standard F
Approximately 40 slower than Standard
14Actel MX Family
- low power consumption
- 5.0V, 3.3V and mixed voltage systems compatible
- design security
15Actel MX Family
- Naming convention
- Six devices in the MX family, vary in number of
equivalent gates - Five speed grades
- A variety of package options and operating
temperature ranges - Same convention used for all Actel FPGAs
16Actel MX Family
1740MX Logic Module
- Same as ACT 1 logic element
- eight-input, one-output logic circuit (702
possibilities) - can implement the four basic logic functions
(NAND, AND, OR and NOR) in gates of two, three,
or four inputs - can also implement a variety of D latches,
exclusivity functions, AND-ORs and OR-ANDs
1842MX C-Module Implementation
- 42MX devices contain three types of logic
modules - combinatorial (C-modules)
- sequential (S-modules)
- decode (D-modules)
- figure at right illustrates the combinatorial
logic module - 766 possibilities
1942MX S-Module Implementation
- S-module implements the same combinatorial logic
function as the C-module while adding a
sequential element. - Sequential element can be configured as either a
D-flip-flop or a transparent latch. - S-module register can be bypassed so that it
implements purely combinatorial (combinational)
logic functions.
S-Module
2042MX S-Module Implementation
- A closer look at the S-Module configurations
21A42MX24 and A42MX36 D-ModuleImplementation
- D-modules are arranged around the periphery of
the device. - D-modules contain wide-decode circuitry,
providing a fast, wide-input AND function similar
to that found in CPLD architectures
22A42MX36 Dual-Port SRAM Block
- The SRAM modules are arranged in 256-bit blocks
that can be configured as 32x8 or 64x4. - SRAM modules can be cascaded together to form
memory spaces of user-definable width and depth. - A42MX36 SRAM modules contain independent read and
write ports.
23From Before Actels ACT Series
- Based on channeled gate array architecture
- Segmented routing tracks
- Interconnect and logic elements in same plane
- Each logic element (labelled L) is a
combination of multiplexers which can be
configured as a multi-input gate
24MX Routing Structure
- MX architecture uses vertical and horizontal
routing tracks to interconnect the logic and I/O
modules. - Routing tracks are metal interconnects that may
be continuous or split into segments. - Varying segment lengths allow the interconnect of
over 90 of design tracks only two antifuse
connections. - Segments can be joined together at the ends using
antifuses to increase their lengths up to the
full length of the track. - All interconnects can be accomplished with a
maximum of four antifuses.
Actels Channeled Routing
25Horizontal Routing
- Horizontal routing tracks span the whole row
length or are divided into multiple segments and
are located in between the rows of modules. - Any segment that spans more than one-third of the
row length is considered a long horizontal
segment. - Within horizontal routing, dedicated routing
tracks are used for global clock networks and for
power and ground tie-off tracks. - Non-dedicated tracks are used for signal nets.
26Vertical Routing
- Another set of routing tracks run vertically.
- There are three types of vertical tracks
- input, output, and long
- Long tracks span the column length of the module,
and can be divided into multiple segments - Each segment in an input track is dedicated to
the input of a particular module - Each segment in an output track is dedicated to
the output of a particular module
27Vertical Routing (continued)
- Long segments are uncommitted and can be assigned
during routing - Each output segment spans four channels
- two above and two below the logic module
- except near the top and bottom of the array,
where edge effects occur - Long vertical tracks contain either one or two
segments
28Antifuse Structures
- An antifuse is a "normally open" structure
- the opposite of a normally closed fuse
- The use of antifuses to implement a programmable
logic device results in - highly testable structures
- efficient programming algorithms
- There are no pre-existing connections
- temporary connections can be made using pass
transistors
29Antifuse Structures (continued)
- These temporary connections can isolate
- individual antifuses to be programmed
- individual circuit structures to be tested
- can be done before and after programming
- For example
- all metal tracks can be tested for continuity and
shorts between adjacent tracks - the functionality of all logic modules can be
verified
30From Before Actel ACT Series I/O Logic
- Simple tristate buffers
- Allowed pins to be used as an input or an output,
or both
ACT 1 I/O Module
ACT 2 I/O Module
ACT 3 I/O Module
31Actel MX Family
- 42MX devices feature Multiplex I/Os and support
5.0V, 3.3V, and mixed 3.3V/5.0V operations - provide the interface between the device pins and
the logic array - modules contain tristate buffers, with input and
output latches that can be configured for input,
output, or bidirectional operation.
42MX I/O Module
32Actel MX Family
- A42MX24 and A42MX36 devices also offer selectable
PCI output drives, enabling 100 compliance with
version 2.1 of the PCI specification - chip-wide PCI fuse
- When the PCI fuse is not programmed, the output
drive is standard.
PCI Output Structure of A42MX24 and A42MX36
Devices
33Actel eX Family
- Low power consumption
- Extremely small chip-scale packages
- Design security
- Nonvolatile single chip
- Up to 100 resource utilization with 100 pin
locking - 2.5V, 3.3V, and 5.0V Mixed Voltage Operation with
5.0V Input Tolerance and 5.0V Drive Strength
34Actel eX Family
- eX naming convention
- three devices in the eX family, vary in number of
equivalent gates and flip-flops - three speed grades
- a variety of package options and operating
temperature ranges - same convention used for all Actel FPGAs
35Actel eX Family
36Actel eX Family
- The eX family architecture uses a
sea-of-modules structure where the entire floor
of the device is covered with a grid of logic
modules with virtually no chip area lost to
interconnect elements or routing. - Interconnection among these logic modules is
achieved using metal-to-metal programmable
antifuse interconnect elements.
37Actel eX Family
- eX family provides two types of logic modules
- the register cell (R-Cell)
- the combinatorial (combinational) cell (C-Cell)
- C-Cell is shown at right
- similar to ACT logic cell of the same name
- inverter input (DB) added
- allows 4000 possible combinational functions to
be implemented
C-Cell
38eX R-Cell
- The R-Cell contains a flip-flop featuring
asynchronous clear, asynchronous preset, and
clock enable (using the S0 and S1 lines) control
signals. - The R-Cell registers feature programmable clock
polarity selectable on a register-by-register
basis. - This provides additional flexibility while
allowing mapping of synthesized functions into
the eX FPGA. - The clock source for the R-Cell can be chosen
from either the hard-wired clock or the routed
clock.
R-Cell
39Actel eX Family
- C-cell and R-cell logic modules are arranged into
horizontal banks called Clusters, each of which
contains two C-cells and one R-cell in a C-R-C
configuration.
- Clusters are further organized into modules
called SuperClusters. - Each SuperCluster is a two-wide grouping of
Clusters.
Cluster Organization
40Actel eX Family Routing Resources
- Clusters and SuperClusters can be connected
through the use of two local routing resources
called FastConnect and DirectConnect.
- This routing architecture reduces the number of
antifuses required to complete a circuit. - DirectConnect is a horizontal routing resource
that provides connections from a C-cell to its
neighboring Rcell in a given SuperCluster.
DirectConnect and FastConnect for SuperClusters
41Actel eX Family Routing Resources
- DirectConnect uses a hardwired signal path
requiring no programmable interconnection. - FastConnect enables horizontal routing between
any two logic modules within a given SuperCluster
and vertical routing with the SuperCluster
immediately below it. - Only one programmable connection is used in a
FastConnect path.
42Actel eX Family
- eX devices have a variety of I/O features, such
as PCI compliance, programmable input threshold
voltage, configurable output slew rate, and
selectable output state during power-up - support for both hot-swapping and cold-sparing
- mixed I/O standards can be set for individual
pins - only allowed with standards that use the same
supply voltage
Simplified I/O Circuitry
43Actel eX Family
- Each I/O on an eX device can be configured as an
input, an output, a tristate output, or a
bidirectional pin - I/O cells in eX devices do not contain embedded
latches or flip-flops and can be inferred
directly from HDL code.
I/O Features
44Actel SX-A and SX Family
- 12,000 to 108,000 usable system gates
- 3.7ns clock-to-output (pin-to-pin)
- 350MHz internal clock frequency
- 66MHz, 64-bit 3.3V/5.0V PCI performance
- Hot swappable I/Os (SX-A)
- Able to run at OC3 to OC192 data rates
- 2.5V, 3.3V, and 5.0V mixed voltage support
- 100 resource utilization with 100 pin locking
- Low power consumption (less than 1W _at_ 200MHz)
- Complete BST/JTAG support
45Actel SX-A and SX Family
- 54SX Family FPGAs
- available in four speed grades
- six package options
- four component densities
- four operating temperature ranges
54SX32A
46Actel SX-A and SX Family
- As with the eX Family, the SX-A and SX Families
use Actels Sea of Modules architecture
47Actel SX-A and SX Family
- the C-cell implements a range of combinatorial
functions up to 5-inputs - inclusion of the DB input and its associated
inverter function dramatically increases the
number of combinatorial functions that can be
implemented in a single module from 800 options
in previous architectures to more than 4,000 in
the SX architecture
C-Cell
48Actel SX-A and SX Family
- the R-cell contains a flip-flop featuring
- asynchronous clear
- asynchronous preset
- clock enable (using the S0 and S1 lines) control
signals - the R-cell registers feature programmable clock
polarity selectable on a register-by-register
basis.
R-Cell
49Actel SX-A and SX Family
- C-cell and R-cell logic modules are arranged in
horizontal banks called Clusters - two types of Clusters
- Type 1 contains two C-cells and one R-cell
- Type 2 contains one C-cell and two R-cells
- SuperCluster 1 is a two-wide grouping of Type 1
clusters. - SuperCluster 2 is a two-wide group containing one
Type 1 cluster and one Type 2 cluster
Cluster Organization
50Actel SX-A and SX Family
- Clusters and SuperClusters can be connected
through the use of two local routing resources
called FastConnect and DirectConnect - This routing architecture reduces the number of
antifuses required to complete a circuit - DirectConnect is a horizontal routing resource
that provides connections from a C-cell to its
neighboring R-cell in a given SuperCluster
DirectConnect and FastConnect for Type 1
SuperClusters
51Actel SX-A and SX Family
- DirectConnect uses a hard-wired signal path
requiring no programmable interconnection to
achieve its fast signal propagation time of less
than 0.1 ns - FastConnect enables horizontal routing between
any two logic modules in a given SuperCluster and
vertical routing with the SuperCluster
immediately below it - Only one programmable connection is used in a
FastConnect path, delivering maximum pin-to-pin
propagation of 0.4 ns
DirectConnect and FastConnect for Type 2
SuperClusters
52Actel SX-A and SX Family
- Each I/O module on an SX device can be configured
as - an input
- an output
- a tristate output
- a bidirectional pin
- Even without the inclusion of dedicated I/O
registers, these I/O modules, in combination with
array registers, can achieve clock-to-out
(pad-to-pad) timing as fast as 3.7 ns. - I/O cells that have embedded latches and
flip-flops require instantiation in HDL code
this is a design complication not encountered in
SX FPGAs.
53Actel SX-A and SX Family
- the SX family locates the routing interconnect
resources between the Metal 2 and Metal 3 layers - this eliminates the channels of routing and
interconnect resources between logic modules - metal-to-metal programmable antifuse interconnect
elements, which are embedded between the M2 and
M3 layers
SX Family Interconnect Elements
54Actel SX-A and SX Family
- SX-A interconnect resources are located between
the top two metal layers - A54SX72A has four layers of metal with the
antifuse between Metal 3 and Metal 4 - A54SX08A, A54SX16A and A54SX32A have three
layers of metal with antifuse between Metal 2 and
Metal 3 - Same as SX
SX-A Family Interconnect Elements
55More on Wednesday