vtr_flow/arch/custom_grid/custom_sbloc.xml - third_party/vtr-verilog-to-routing - Git at Google

 <!--
   Flagship Heterogeneous Architecture with Carry Chains for VTR 7.0.

   - 40 nm technology
   - General purpose logic block:
     K = 6, N = 10, fracturable 6 LUTs (can operate as one 6-LUT or two 5-LUTs with all 5 inputs shared)
     with optionally registered outputs
     Each 5-LUT has an arithemtic mode that converts it to a single-bit adder with both inputs driven by 4-LUTs (both 4-LUTs share all 4 inputs)
     Carry chain links to vertically adjacent logic blocks
   - Memory size 32 Kbits, memory aspect ratios vary from a data width of 1 to data width of 64.
     Height = 6, found on every (8n+2)th column
   - Multiplier modes: one 36x36, two 18x18, each 18x18 can also operate as two 9x9.
     Height = 4, found on every (8n+6)th column
   - Routing architecture: L = 4, fc_in = 0.15, Fc_out = 0.1

   Details on Modelling:

   The electrical design of the architecture described here is NOT from an
   optimized, SPICED architecture.  Instead, we attempt to create a reasonable
   architecture file by using an existing commercial FPGA to approximate the area,
   delay, and power of the underlying components. This is combined with a reasonable 40 nm
   model of wiring and circuit design for low-level routing components, where available.
   The resulting architecture has delays that roughly match a commercial 40 nm FPGA, but also
   has wiring electrical parameters that allow the wire lengths and switch patterns to be
   modified and you will still get reasonable delay results for the new architecture.
   The following describes, in detail, how we obtained the various electrical values for this
   architecture.

   Rmin for nmos and pmos, routing buffer sizes, and I/O pad delays are from the ifar
   architecture created by Ian Kuon: K06 N10 45nm fc 0.15 area-delay optimized architecture.
   (n10k06l04.fc15.area1delay1.cmos45nm.bptm.cmos45nm.xml)
   This routing architecture was optimized for 45 nm, and we have scaled it linearly to 40 nm to
   match the overall target (a 40 nm FPGA).

   We obtain maximum delay numbers by measuring delays of routing, soft logic blocks,
   memories, and multipliers from test circuits on a Stratix IV GX device
   (EP4SGX230DF29C2X, i.e. fastest speed grade). Minimum delay values are calculated based on the
   ratios between maximum and minimum values in Stratix IV GX device. For routing, we took the
   average delay of H4 and V4 wires.  Rmetal and Cmetal values for the routing wires were obtained
   from work done by Charles Chiasson. We use a 96 nm half-pitch (corresponding to mid-level metal
   stack 40 nm routing) and take the R and C data from the ITRS roadmap.

   For the general purpose logic block, we assume that the area and delays of the Stratix IV
   crossbar is close enough to the crossbar modelled here.  We use 40 inputs and 20 feedback lines in
   the cluster and a full crossbar, leading to 53:1 multiplexers in front of each BLE input.
   Stratix IV uses 52 inputs and 20 feedback lines, but only a half-populated crossbar, leading to
   36:1 multiplexers.  We require 60 such multiplexers, while Stratix IV requires 88 for its more
   complex fracturable BLEs + the extra control signals. We justify this rough approximation as follows:
   The Stratix IV crossbar has more inputs (72 vs. 60) and
   outputs (88 vs. 60) than our full crossbar which should increase its area and delay, but the
   Stratix IV crossbar is also 50% sparse (each mux is 36:1 instead of 53:1) which should reduce its
   area and delay.  The total number of crossbar switch points is roughly similar between the two
   architectures (3160 for SIV and 3600 for the academic architecture below), so we use the area
   & delay of the Stratix IV crossbar as a rough approximation of our crossbar.

   For LUTs, we include LUT
   delays measured from Stratix IV which is dependant on the input used (ie. some
   LUT inputs are faster than others).  The CAD tools at the time of VTR 7 does
   not consider differences in LUT input delays.

   Adder delays obtained as approximate values from a Stratix IV EP4SE230F29C3 device.
   Delay obtained by compiling a 256 bit adder (registered inputs and outputs,
   all pins except clock virtual) then measuring the delays in chip-planner,
   sumout delay = 0.271ns to 0.348 ns, intra-block carry delay = 0.011 ns,
   inter-block carry delay = 0.327 ns.  Given this data, I will approximate
   sumout 0.3 ns, intra-block carry-delay = 0.01 ns, and
   inter-block carry-delay = 0.16 ns (since Altera inter-block carry delay has
   overhead that we don't have, I'll approximate the delay of a simpler chain at
   one half what they have.  This is very rough, anything from 0.01ns to 0.327ns
   can be justified).

   Logic block area numbers obtained by scaling overall tile area of a 65nm
   Stratix III device, (as given in Wong, Betz and Rose, FPGA 2011) to 40 nm, then subtracting out
   routing area at a channel width of 300. We use a channel width of 300 because it can route
   all the VTR 6.0 benchmark circuits with an approximately 20% safety margin, and is also close to the
   total channel width of Stratix IV. Hence this channel width is close to the commercial practice of
   choosing a width that provides high routability. The architecture can be routed at different channel
   widths, but we estimate the tile size and hence the physical length of routing wires assuming
   a channel width of 300.

   Sanity checks employed:
     1.  We confirmed the routing buffer delay is ~1/3rd of total routing delay at L = 4. This matches
         common electrical design.


   Authors: Jason Luu, Jeff Goeders, Vaughn Betz
 -->
 <architecture>
   <!--
        ODIN II specific config begins
        Describes the types of user-specified netlist blocks (in blif, this corresponds to
        ".model [type_of_block]") that this architecture supports.

        Note: Basic LUTs, I/Os, and flip-flops are not included here as there are
        already special structures in blif (.names, .input, .output, and .latch)
        that describe them.
   -->
   <models>
     <model name="multiply">
       <input_ports>
         <port name="a" combinational_sink_ports="out"/>
         <port name="b" combinational_sink_ports="out"/>
       </input_ports>
       <output_ports>
         <port name="out"/>
       </output_ports>
     </model>
     <model name="single_port_ram">
       <input_ports>
         <port name="we" clock="clk"/>
         <!-- control -->
         <port name="addr" clock="clk"/>
         <!-- address lines -->
         <port name="data" clock="clk"/>
         <!-- data lines can be broken down into smaller bit widths minimum size 1 -->
         <port name="clk" is_clock="1"/>
         <!-- memories are often clocked -->
       </input_ports>
       <output_ports>
         <port name="out" clock="clk"/>
         <!-- output can be broken down into smaller bit widths minimum size 1 -->
       </output_ports>
     </model>
     <model name="dual_port_ram">
       <input_ports>
         <port name="we1" clock="clk"/>
         <!-- write enable -->
         <port name="we2" clock="clk"/>
         <!-- write enable -->
         <port name="addr1" clock="clk"/>
         <!-- address lines -->
         <port name="addr2" clock="clk"/>
         <!-- address lines -->
         <port name="data1" clock="clk"/>
         <!-- data lines can be broken down into smaller bit widths minimum size 1 -->
         <port name="data2" clock="clk"/>
         <!-- data lines can be broken down into smaller bit widths minimum size 1 -->
         <port name="clk" is_clock="1"/>
         <!-- memories are often clocked -->
       </input_ports>
       <output_ports>
         <port name="out1" clock="clk"/>
         <!-- output can be broken down into smaller bit widths minimum size 1 -->
         <port name="out2" clock="clk"/>
         <!-- output can be broken down into smaller bit widths minimum size 1 -->
       </output_ports>
     </model>
     <model name="adder">
       <input_ports>
         <port name="a" combinational_sink_ports="sumout cout"/>
         <port name="b" combinational_sink_ports="sumout cout"/>
         <port name="cin" combinational_sink_ports="sumout cout"/>
       </input_ports>
       <output_ports>
         <port name="cout"/>
         <port name="sumout"/>
       </output_ports>
     </model>
   </models>
   <tiles>
     <tile name="io" capacity="8" area="0">
       <equivalent_sites>
         <site pb_type="io"/>
       </equivalent_sites>
       <input name="outpad" num_pins="1"/>
       <output name="inpad" num_pins="1"/>
       <clock name="clock" num_pins="1"/>
       <fc in_type="frac" in_val="0.15" out_type="frac" out_val="0.10"/>
       <pinlocations pattern="custom">
         <loc side="left">io.outpad io.inpad io.clock</loc>
         <loc side="top">io.outpad io.inpad io.clock</loc>
         <loc side="right">io.outpad io.inpad io.clock</loc>
         <loc side="bottom">io.outpad io.inpad io.clock</loc>
       </pinlocations>
     </tile>
     <tile name="clb" area="53894">
       <equivalent_sites>
         <site pb_type="clb"/>
       </equivalent_sites>
       <input name="I" num_pins="40" equivalent="full"/>
       <input name="cin" num_pins="1"/>
       <output name="O" num_pins="20" equivalent="none"/>
       <output name="cout" num_pins="1"/>
       <clock name="clk" num_pins="1"/>
       <fc in_type="frac" in_val="0.15" out_type="frac" out_val="0.10">
         <fc_override port_name="cin" fc_type="frac" fc_val="0"/>
         <fc_override port_name="cout" fc_type="frac" fc_val="0"/>
       </fc>
       <pinlocations pattern="spread"/>
     </tile>
     <tile name="mult_36" width="3" height="3" area="396000">
       <equivalent_sites>
         <site pb_type="mult_36"/>
       </equivalent_sites>
       <input name="a" num_pins="36"/>
       <input name="b" num_pins="36"/>
       <output name="out" num_pins="72"/>
       <fc in_type="frac" in_val="0.15" out_type="frac" out_val="0.10"/>
       <pinlocations pattern="spread"/>
       <switchblock_locations pattern="custom">
         <!-- Full outside of block -->
         <sb_loc type="full" xoffset="0" yoffset="2"/>
         <sb_loc type="full" xoffset="1" yoffset="2"/>
         <sb_loc type="full" xoffset="2" yoffset="2"/>
         <sb_loc type="full" xoffset="2" yoffset="1"/>
         <sb_loc type="full" xoffset="2" yoffset="0"/>
         <!-- Test all the sb types on internal locations -->
         <sb_loc type="full" xoffset="0" yoffset="0"/>
         <sb_loc type="straight" xoffset="0" yoffset="1"/>
         <sb_loc type="turns" xoffset="1" yoffset="0"/>
         <sb_loc type="none" xoffset="1" yoffset="1"/>
       </switchblock_locations>
     </tile>
     <tile name="memory" width="2" height="3" area="548000">
       <equivalent_sites>
         <site pb_type="memory"/>
       </equivalent_sites>
       <input name="addr1" num_pins="15"/>
       <input name="addr2" num_pins="15"/>
       <input name="data" num_pins="64"/>
       <input name="we1" num_pins="1"/>
       <input name="we2" num_pins="1"/>
       <output name="out" num_pins="64"/>
       <clock name="clk" num_pins="1"/>
       <fc in_type="frac" in_val="0.15" out_type="frac" out_val="0.10"/>
       <pinlocations pattern="spread"/>
       <switchblock_locations pattern="external"/>
     </tile>
   </tiles>
   <!-- ODIN II specific config ends -->
   <!-- Physical descriptions begin -->
   <layout>
     <auto_layout>
       <!--Perimeter of 'io' blocks with 'EMPTY' blocks at corners-->
       <perimeter type="io" priority="100"/>
       <corners type="EMPTY" priority="101"/>
       <!--Fill with 'clb'-->
       <fill type="clb" priority="10"/>
       <!--Column of 'mult_36' with 'EMPTY' blocks wherever a 'mult_36' does not fit. Vertical offset by 1 for perimeter.-->
       <col type="mult_36" startx="6" starty="1" repeatx="8" priority="20"/>
       <col type="EMPTY" startx="6" repeatx="8" starty="1" priority="19"/>
       <!--Column of 'memory' with 'EMPTY' blocks wherever a 'memory' does not fit. Vertical offset by 1 for perimeter.-->
       <col type="memory" startx="2" starty="1" repeatx="8" priority="20"/>
       <col type="EMPTY" startx="2" repeatx="8" starty="1" priority="19"/>
     </auto_layout>
   </layout>
   <device>
     <!-- VB & JL: Using Ian Kuon's transistor sizing and drive strength data for routing, at 40 nm. Ian used BPTM
 			     models. We are modifying the delay values however, to include metal C and R, which allows more architecture
 			     experimentation. We are also modifying the relative resistance of PMOS to be 1.8x that of NMOS
 			     (vs. Ian's 3x) as 1.8x lines up with Jeff G's data from a 45 nm process (and is more typical of
 			     45 nm in general). I'm upping the Rmin_nmos from Ian's just over 6k to nearly 9k, and dropping
 			     RminW_pmos from 18k to 16k to hit this 1.8x ratio, while keeping the delays of buffers approximately
 			     lined up with Stratix IV.
 			     We are using Jeff G.'s capacitance data for 45 nm (in tech/ptm_45nm).
 			     Jeff's tables list C in for transistors with widths in multiples of the minimum feature size (45 nm).
 			     The minimum contactable transistor is 2.5 * 45 nm, so I need to multiply drive strength sizes in this file
 	                     by 2.5x when looking up in Jeff's tables.
 			     The delay values are lined up with Stratix IV, which has an architecture similar to this
 			     proposed FPGA, and which is also 40 nm
 			     C_ipin_cblock: input capacitance of a track buffer, which VPR assumes is a single-stage
 			     4x minimum drive strength buffer. -->
     <sizing R_minW_nmos="8926" R_minW_pmos="16067"/>
     <!-- The grid_logic_tile_area below will be used for all blocks that do not explicitly set their own (non-routing)
      	  area; set to 0 since we explicitly set the area of all blocks currently in this architecture file.
 	    -->
     <area grid_logic_tile_area="0"/>
     <chan_width_distr>
       <x distr="uniform" peak="1.000000"/>
       <y distr="uniform" peak="1.000000"/>
     </chan_width_distr>
     <switch_block type="wilton" fs="3"/>
     <connection_block input_switch_name="ipin_cblock"/>
   </device>
   <switchlist>
     <!-- VB: the mux_trans_size and buf_size data below is in minimum width transistor *areas*, assuming the purple
            book area formula. This means the mux transistors are about 5x minimum drive strength.
            We assume the first stage of the buffer is 3x min drive strength to be reasonable given the large
            mux transistors, and this gives a reasonable stage ratio of a bit over 5x to the second stage. We assume
            the n and p transistors in the first stage are equal-sized to lower the buffer trip point, since it's fed
            by a pass transistor mux. We can then reverse engineer the buffer second stage to hit the specified
            buf_size (really buffer area) - 16.2x minimum drive nmos and 1.8*16.2 = 29.2x minimum drive.
            I then took the data from Jeff G.'s PTM modeling of 45 nm to get the Cin (gate of first stage) and Cout
            (diff of second stage) listed below.  Jeff's models are in tech/ptm_45nm, and are in min feature multiples.
            The minimum contactable transistor is 2.5 * 45 nm, so I need to multiply the drive strength sizes above by
            2.5x when looking up in Jeff's tables.
            Finally, we choose a switch delay (58 ps) that leads to length 4 wires having a delay equal to that of SIV of 126 ps.
            This also leads to the switch being 46% of the total wire delay, which is reasonable. -->
     <switch type="mux" name="0" R="551" Cin=".77e-15" Cout="4e-15" Tdel="58e-12" mux_trans_size="2.630740" buf_size="27.645901"/>
     <!--switch ipin_cblock resistance set to yeild for 4x minimum drive strength buffer-->
     <switch type="mux" name="ipin_cblock" R="2231.5" Cout="0." Cin="1.47e-15" Tdel="7.247000e-11" mux_trans_size="1.222260" buf_size="auto"/>
   </switchlist>
   <segmentlist>
     <!--- VB & JL: using ITRS metal stack data, 96 nm half pitch wires, which are intermediate metal width/space.
              With the 96 nm half pitch, such wires would take 60 um of height, vs. a 90 nm high (approximated as square) Stratix IV tile so this seems
              reasonable. Using a tile length of 90 nm, corresponding to the length of a Stratix IV tile if it were square. -->
     <segment freq="1.000000" length="4" type="unidir" Rmetal="101" Cmetal="22.5e-15">
       <mux name="0"/>
       <sb type="pattern">1 1 1 1 1</sb>
       <cb type="pattern">1 1 1 1</cb>
     </segment>
   </segmentlist>
   <directlist>
     <direct name="adder_carry" from_pin="clb.cout" to_pin="clb.cin" x_offset="0" y_offset="-1" z_offset="0"/>
   </directlist>
   <complexblocklist>
     <!-- Define I/O pads begin -->
     <!-- Capacity is a unique property of I/Os, it is the maximum number of I/Os that can be placed at the same (X,Y) location on the FPGA -->
     <!-- Not sure of the area of an I/O (varies widely), and it's not relevant to the design of the FPGA core, so we're setting it to 0. -->
     <pb_type name="io">
       <input name="outpad" num_pins="1"/>
       <output name="inpad" num_pins="1"/>
       <clock name="clock" num_pins="1"/>
       <!-- IOs can operate as either inputs or outputs.
 	     Maximum delays below come from Ian Kuon. They are small, so they should be interpreted as
 	     the delays to and from registers in the I/O (and generally I/Os are registered
 	     today and that is when you timing analyze them.

 		 Minimum delays are retrieved using a ratio of maximum and minimum times as seen in Quartus II
 		 in Stratix IV. The ratio of minimum value/maximum value is as follows:
 			inpad delay:  0.9239
 			outpad delay: 0.9545

 	     -->
       <mode name="inpad">
         <pb_type name="inpad" blif_model=".input" num_pb="1">
           <output name="inpad" num_pins="1"/>
         </pb_type>
         <interconnect>
           <direct name="inpad" input="inpad.inpad" output="io.inpad">
             <delay_constant max="4.243e-11" min="3.92e-11" in_port="inpad.inpad" out_port="io.inpad"/>
           </direct>
         </interconnect>
       </mode>
       <mode name="outpad">
         <pb_type name="outpad" blif_model=".output" num_pb="1">
           <input name="outpad" num_pins="1"/>
         </pb_type>
         <interconnect>
           <direct name="outpad" input="io.outpad" output="outpad.outpad">
             <delay_constant max="1.394e-11" min="1.331e-11" in_port="io.outpad" out_port="outpad.outpad"/>
           </direct>
         </interconnect>
       </mode>
       <!-- Every input pin is driven by 15% of the tracks in a channel, every output pin is driven by 10% of the tracks in a channel -->
       <!-- IOs go on the periphery of the FPGA, for consistency,
           make it physically equivalent on all sides so that only one definition of I/Os is needed.
           If I do not make a physically equivalent definition, then I need to define 4 different I/Os, one for each side of the FPGA
         -->
       <!-- Place I/Os on the sides of the FPGA -->
       <power method="ignore"/>
     </pb_type>
     <!-- Define I/O pads ends -->
     <!-- Define general purpose logic block (CLB) begin -->
     <!--- Area calculation: Total Stratix IV tile area is about 8100 um^2, and a minimum width transistor
 	   area is 60 L^2 yields a tile area of 84375 MWTAs.
 	   Routing at W=300 is 30481 MWTAs, leaving us with a total of 53000 MWTAs for logic block area
 	   This means that only 37% of our area is in the general routing, and 63% is inside the logic
 	   block. Note that the crossbar / local interconnect is considered part of the logic block
 	   area in this analysis. That is a lower proportion of of routing area than most academics
 	   assume, but note that the total routing area really includes the crossbar, which would push
 	   routing area up significantly, we estimate into the ~70% range.
 	   -->
     <pb_type name="clb">
       <input name="I" num_pins="40" equivalent="full"/>
       <input name="cin" num_pins="1"/>
       <output name="O" num_pins="20" equivalent="none"/>
       <output name="cout" num_pins="1"/>
       <clock name="clk" num_pins="1"/>
       <!-- Describe fracturable logic element.
              Each fracturable logic element has a 6-LUT that can alternatively operate as two 5-LUTs with shared inputs.
              The outputs of the fracturable logic element can be optionally registered
         -->
       <pb_type name="fle" num_pb="10">
         <input name="in" num_pins="6"/>
         <input name="cin" num_pins="1"/>
         <output name="out" num_pins="2"/>
         <output name="cout" num_pins="1"/>
         <clock name="clk" num_pins="1"/>
         <mode name="n2_lut5">
           <pb_type name="lut5inter" num_pb="1">
             <input name="in" num_pins="5"/>
             <input name="cin" num_pins="1"/>
             <output name="out" num_pins="2"/>
             <output name="cout" num_pins="1"/>
             <clock name="clk" num_pins="1"/>
             <pb_type name="ble5" num_pb="2">
               <input name="in" num_pins="5"/>
               <input name="cin" num_pins="1"/>
               <output name="out" num_pins="1"/>
               <output name="cout" num_pins="1"/>
               <clock name="clk" num_pins="1"/>
               <mode name="blut5">
                 <pb_type name="flut5" num_pb="1">
                   <input name="in" num_pins="5"/>
                   <output name="out" num_pins="1"/>
                   <clock name="clk" num_pins="1"/>
                   <!-- Regular LUT mode -->
                   <pb_type name="lut5" blif_model=".names" num_pb="1" class="lut">
                     <input name="in" num_pins="5" port_class="lut_in"/>
                     <output name="out" num_pins="1" port_class="lut_out"/>
                     <!-- LUT timing using delay matrix -->
                     <!-- These are the physical maximum delay inputs on a Stratix IV LUT but because VPR cannot do LUT rebalancing,
                            we instead take the average of these numbers to get more stable results
                         82e-12
                         173e-12
                         261e-12
                         263e-12
                         398e-12
 							The minimum delay/maximum delay ratio is 0.7395 in QII on Stratix IV. Hence, the minimum delay
 						is 0.7295 * the average of the maximum numbers
                         -->
                     <delay_matrix type="max" in_port="lut5.in" out_port="lut5.out">
                         235e-12
                         235e-12
                         235e-12
                         235e-12
                         235e-12
                       </delay_matrix>
                     <delay_matrix type="min" in_port="lut5.in" out_port="lut5.out">
                         174e-12
                         174e-12
                         174e-12
                         174e-12
                         174e-12
                       </delay_matrix>
                   </pb_type>
                   <pb_type name="ff" blif_model=".latch" num_pb="1" class="flipflop">
                     <input name="D" num_pins="1" port_class="D"/>
                     <output name="Q" num_pins="1" port_class="Q"/>
                     <clock name="clk" num_pins="1" port_class="clock"/>
                     <T_setup value="66e-12" port="ff.D" clock="clk"/>
                     <T_hold value="37e-12" port="ff.D" clock="clk"/>
                     <T_clock_to_Q max="124e-12" min="60e-12" port="ff.Q" clock="clk"/>
                   </pb_type>
                   <interconnect>
                     <direct name="direct1" input="flut5.in" output="lut5.in"/>
                     <direct name="direct2" input="lut5.out" output="ff.D">
                       <pack_pattern name="ble5" in_port="lut5.out" out_port="ff.D"/>
                     </direct>
                     <direct name="direct3" input="flut5.clk" output="ff.clk"/>
                     <mux name="mux1" input="ff.Q lut5.out" output="flut5.out">
                       <delay_constant max="25e-12" min="24e-12" in_port="lut5.out" out_port="flut5.out"/>
                       <delay_constant max="45e-12" min="27e-12" in_port="ff.Q" out_port="flut5.out"/>
                     </mux>
                   </interconnect>
                 </pb_type>
                 <interconnect>
                   <direct name="direct1" input="ble5.in" output="flut5.in"/>
                   <direct name="direct2" input="ble5.clk" output="flut5.clk"/>
                   <direct name="direct3" input="flut5.out" output="ble5.out"/>
                 </interconnect>
               </mode>
               <mode name="arithmetic">
                 <pb_type name="arithmetic" num_pb="1">
                   <input name="in" num_pins="4"/>
                   <input name="cin" num_pins="1"/>
                   <output name="out" num_pins="1"/>
                   <output name="cout" num_pins="1"/>
                   <clock name="clk" num_pins="1"/>
                   <!-- Special dual-LUT mode that drives adder only -->
                   <pb_type name="lut4" blif_model=".names" num_pb="2" class="lut">
                     <input name="in" num_pins="4" port_class="lut_in"/>
                     <output name="out" num_pins="1" port_class="lut_out"/>
                     <!-- LUT timing using delay matrix -->
                     <!-- These are the physical delay inputs on a Stratix IV LUT but because VPR cannot do LUT rebalancing,
                              we instead take the average of these numbers to get more stable results
                         82e-12
                         173e-12
                         261e-12
                         263e-12
 							The minimum delay/maximum delay ratio is 0.7395 in QII on Stratix IV. Hence, the minimum delay
 						is 0.7295 * the average of the maximum numbers
                         -->
                     <delay_matrix type="max" in_port="lut4.in" out_port="lut4.out">
                         195e-12
                         195e-12
                         195e-12
                         195e-12
                       </delay_matrix>
                     <delay_matrix type="min" in_port="lut4.in" out_port="lut4.out">
                         144e-12
                         144e-12
                         144e-12
                         144e-12
                       </delay_matrix>
                   </pb_type>
                   <!-- The ratio between minimum and maximum delays in StratixIV for data ports to sumout
 							is 0.6809 and cin to sumout is 0.6969-->
                   <pb_type name="adder" blif_model=".subckt adder" num_pb="1">
                     <input name="a" num_pins="1"/>
                     <input name="b" num_pins="1"/>
                     <input name="cin" num_pins="1"/>
                     <output name="cout" num_pins="1"/>
                     <output name="sumout" num_pins="1"/>
                     <delay_constant max="0.3e-9" min="0.2043e-9" in_port="adder.a" out_port="adder.sumout"/>
                     <delay_constant max="0.3e-9" min="0.2043e-9" in_port="adder.b" out_port="adder.sumout"/>
                     <delay_constant max="0.3e-9" min="0.2043e-9" in_port="adder.cin" out_port="adder.sumout"/>
                     <delay_constant max="0.3e-9" min="0.2043e-9" in_port="adder.a" out_port="adder.cout"/>
                     <delay_constant max="0.3e-9" min="0.2043e-9" in_port="adder.b" out_port="adder.cout"/>
                     <delay_constant max="0.01e-9" min="6.9797e-12" in_port="adder.cin" out_port="adder.cout"/>
                   </pb_type>
                   <pb_type name="ff" blif_model=".latch" num_pb="1" class="flipflop">
                     <input name="D" num_pins="1" port_class="D"/>
                     <output name="Q" num_pins="1" port_class="Q"/>
                     <clock name="clk" num_pins="1" port_class="clock"/>
                     <T_setup value="66e-12" port="ff.D" clock="clk"/>
                     <T_hold value="37e-12" port="ff.D" clock="clk"/>
                     <T_clock_to_Q max="124e-12" min="60e-12" port="ff.Q" clock="clk"/>
                   </pb_type>
                   <interconnect>
                     <direct name="clock" input="arithmetic.clk" output="ff.clk"/>
                     <direct name="lut_in1" input="arithmetic.in[3:0]" output="lut4[0:0].in[3:0]"/>
                     <direct name="lut_in2" input="arithmetic.in[3:0]" output="lut4[1:1].in[3:0]"/>
                     <direct name="lut_to_add1" input="lut4[0:0].out" output="adder.a">
                       </direct>
                     <direct name="lut_to_add2" input="lut4[1:1].out" output="adder.b">
                       </direct>
                     <direct name="add_to_ff" input="adder.sumout" output="ff.D">
                       <pack_pattern name="chain" in_port="adder.sumout" out_port="ff.D"/>
                     </direct>
                     <direct name="carry_in" input="arithmetic.cin" output="adder.cin">
                       <pack_pattern name="chain" in_port="arithmetic.cin" out_port="adder.cin"/>
                     </direct>
                     <direct name="carry_out" input="adder.cout" output="arithmetic.cout">
                       <pack_pattern name="chain" in_port="adder.cout" out_port="arithmetic.cout"/>
                     </direct>
                     <mux name="sumout" input="ff.Q adder.sumout" output="arithmetic.out">
                       <delay_constant max="25e-12" min="24e-12" in_port="adder.sumout" out_port="arithmetic.out"/>
                       <delay_constant max="45e-12" min="27e-12" in_port="ff.Q" out_port="arithmetic.out"/>
                     </mux>
                   </interconnect>
                 </pb_type>
                 <interconnect>
                   <direct name="direct1" input="ble5.in[3:0]" output="arithmetic.in"/>
                   <direct name="carry_in" input="ble5.cin" output="arithmetic.cin">
                     <pack_pattern name="chain" in_port="ble5.cin" out_port="arithmetic.cin"/>
                   </direct>
                   <direct name="carry_out" input="arithmetic.cout" output="ble5.cout">
                     <pack_pattern name="chain" in_port="arithmetic.cout" out_port="ble5.cout"/>
                   </direct>
                   <direct name="direct2" input="ble5.clk" output="arithmetic.clk"/>
                   <direct name="direct3" input="arithmetic.out" output="ble5.out"/>
                 </interconnect>
               </mode>
             </pb_type>
             <interconnect>
               <direct name="direct1" input="lut5inter.in" output="ble5[0:0].in"/>
               <direct name="direct2" input="lut5inter.in" output="ble5[1:1].in"/>
               <direct name="direct3" input="ble5[1:0].out" output="lut5inter.out"/>
               <direct name="carry_in" input="lut5inter.cin" output="ble5[0:0].cin">
                 <pack_pattern name="chain" in_port="lut5inter.cin" out_port="ble5[0:0].cin"/>
               </direct>
               <direct name="carry_out" input="ble5[1:1].cout" output="lut5inter.cout">
                 <pack_pattern name="chain" in_port="ble5[1:1].cout" out_port="lut5inter.cout"/>
               </direct>
               <direct name="carry_link" input="ble5[0:0].cout" output="ble5[1:1].cin">
                 <pack_pattern name="chain" in_port="ble5[0:0].cout" out_port="ble5[1:1].cout"/>
               </direct>
               <complete name="complete1" input="lut5inter.clk" output="ble5[1:0].clk"/>
             </interconnect>
           </pb_type>
           <interconnect>
             <direct name="direct1" input="fle.in[4:0]" output="lut5inter.in"/>
             <direct name="direct2" input="lut5inter.out" output="fle.out"/>
             <direct name="direct3" input="fle.clk" output="lut5inter.clk"/>
             <direct name="carry_in" input="fle.cin" output="lut5inter.cin">
               <pack_pattern name="chain" in_port="fle.cin" out_port="lut5inter.cin"/>
             </direct>
             <direct name="carry_out" input="lut5inter.cout" output="fle.cout">
               <pack_pattern name="chain" in_port="lut5inter.cout" out_port="fle.cout"/>
             </direct>
           </interconnect>
         </mode>
         <!-- n2_lut5 -->
         <mode name="n1_lut6">
           <pb_type name="ble6" num_pb="1">
             <input name="in" num_pins="6"/>
             <output name="out" num_pins="1"/>
             <clock name="clk" num_pins="1"/>
             <pb_type name="lut6" blif_model=".names" num_pb="1" class="lut">
               <input name="in" num_pins="6" port_class="lut_in"/>
               <output name="out" num_pins="1" port_class="lut_out"/>
               <!-- LUT timing using delay matrix -->
               <!-- These are the physical delay inputs on a Stratix IV LUT but because VPR cannot do LUT rebalancing,
                        we instead take the average of these numbers to get more stable results
                   82e-12
                   173e-12
                   261e-12
                   263e-12
                   398e-12
                   397e-12

 					The minimum delay/maximum delay ratio is 0.7395 in QII on Stratix IV. Hence, the minimum delay
 						is 0.7295 * the average of the maximum numbers
                   -->
               <delay_matrix type="max" in_port="lut6.in" out_port="lut6.out">
                   261e-12
                   261e-12
                   261e-12
                   261e-12
                   261e-12
                   261e-12
                 </delay_matrix>
               <delay_matrix type="min" in_port="lut6.in" out_port="lut6.out">
                   174e-12
                   174e-12
                   174e-12
                   174e-12
                   174e-12
                   174e-12
                 </delay_matrix>
             </pb_type>
             <pb_type name="ff" blif_model=".latch" num_pb="1" class="flipflop">
               <input name="D" num_pins="1" port_class="D"/>
               <output name="Q" num_pins="1" port_class="Q"/>
               <clock name="clk" num_pins="1" port_class="clock"/>
               <T_setup value="66e-12" port="ff.D" clock="clk"/>
               <T_setup value="66e-12" port="ff.D" clock="clk"/>
               <T_hold value="37e-12" port="ff.D" clock="clk"/>
               <T_hold value="37e-12" port="ff.D" clock="clk"/>
               <T_clock_to_Q max="124e-12" min="60e-12" port="ff.Q" clock="clk"/>
             </pb_type>
             <interconnect>
               <direct name="direct1" input="ble6.in" output="lut6[0:0].in"/>
               <direct name="direct2" input="lut6.out" output="ff.D">
                 <pack_pattern name="ble6" in_port="lut6.out" out_port="ff.D"/>
               </direct>
               <direct name="direct3" input="ble6.clk" output="ff.clk"/>
               <mux name="mux1" input="ff.Q lut6.out" output="ble6.out">
                 <delay_constant max="25e-12" min="24e-12" in_port="lut6.out" out_port="ble6.out"/>
                 <delay_constant max="45e-12" min="27e-12" in_port="ff.Q" out_port="ble6.out"/>
               </mux>
             </interconnect>
           </pb_type>
           <interconnect>
             <direct name="direct1" input="fle.in" output="ble6.in"/>
             <direct name="direct2" input="ble6.out" output="fle.out[0:0]"/>
             <direct name="direct3" input="fle.clk" output="ble6.clk"/>
           </interconnect>
         </mode>
         <!-- n1_lut6 -->
       </pb_type>
       <interconnect>
         <!-- We use a full crossbar to get logical equivalence at inputs of CLB
            The delays below come from Stratix IV. the delay through a connection block
            input mux + the crossbar in Stratix IV is 167 ps. We already have a 72 ps
            delay on the connection block input mux (modeled by Ian Kuon), so the remaining
            delay within the crossbar is 95 ps.
 		   For the minimum delays, we have the delay through a connection block input mux +
 		   the crossbar in Stratix IV is 144. Subtracting the 72 ps leaves 72 ps remaining.
            The max delays of cluster feedbacks in Stratix IV is 100 ps, when driven by a LUT.
            Since all our outputs LUT outputs go to a BLE output, and have a delay of
            25 ps to do so, we subtract 25 ps from the 100 ps delay of a feedback
            to get the part that should be marked on the crossbar. For the minimum delay,
   		   the value in Stratix IV is 93 ps, subtracting the 24 ps leaves 69 ps.-->
         <complete name="crossbar" input="clb.I fle[9:0].out" output="fle[9:0].in">
           <delay_constant max="95e-12" min="72e-12" in_port="clb.I" out_port="fle[9:0].in"/>
           <delay_constant max="75e-12" min="69e-12" in_port="fle[9:0].out" out_port="fle[9:0].in"/>
         </complete>
         <complete name="clks" input="clb.clk" output="fle[9:0].clk">
           </complete>
         <!-- This way of specifying direct connection to clb outputs is important because this architecture uses automatic spreading of opins.
                  By grouping to output pins in this fashion, if a logic block is completely filled by 6-LUTs,
                  then the outputs those 6-LUTs take get evenly distributed across all four sides of the CLB instead of clumped on two sides (which is what happens with a more
                  naive specification).
           -->
         <direct name="clbouts1" input="fle[9:0].out[0:0]" output="clb.O[9:0]"/>
         <direct name="clbouts2" input="fle[9:0].out[1:1]" output="clb.O[19:10]"/>
         <!-- Carry chain links -->
         <direct name="carry_in" input="clb.cin" output="fle[0:0].cin">
           <!-- Put all inter-block carry chain delay on this one edge -->
           <delay_constant max="0.16e-9" min="0.11e-9" in_port="clb.cin" out_port="fle[0:0].cin"/>
           <pack_pattern name="chain" in_port="clb.cin" out_port="fle[0:0].cin"/>
         </direct>
         <direct name="carry_out" input="fle[9:9].cout" output="clb.cout">
           <pack_pattern name="chain" in_port="fle[9:9].cout" out_port="clb.cout"/>
         </direct>
         <direct name="carry_link" input="fle[8:0].cout" output="fle[9:1].cin">
           <pack_pattern name="chain" in_port="fle[8:0].cout" out_port="fle[9:1].cin"/>
         </direct>
       </interconnect>
     </pb_type>
     <!-- Define general purpose logic block (CLB) ends -->
     <!-- Define fracturable multiplier begin -->
     <!-- This multiplier can operate as a 36x36 multiplier that can fracture to two 18x18 multipliers each of which can further fracture to two 9x9 multipliers
 	   For delay modelling, the 36x36 DSP multiplier in Stratix IV has a maximum delay of 1.523 ns + 1.93 ns
 	    = 3.45 ns. The average difference between the maximum and minimum values of a dsp mac out in 36 bit multiply
 		mode is 0.51. Hence, the minimum delay is modeled as 0.776 ns + 0.984 ns = 1.760 ns.
  		The 18x18 mode doesn't need to sum four 18x18 multipliers, so it is a bit
 	   faster: 1.523 ns for the multiplier, and 1.09 ns for the multiplier output block.
 	    For the input and output interconnect delays, unlike Stratix IV, we don't
 	   have any routing/logic flexibility (crossbars) at the inputs.  There is some output muxing
 	   in Stratix IV and this architecture to select which multiplier outputs should go out (e.g.
 	   9x9 outputs, 18x18 or 36x36) so those are very close between the two architectures.
 	   We take the conservative (slightly pessimistic)
            approach modelling the input as the same as the Stratix IV input delay and the output delay the same as the Stratix IV DSP out delay.

 	   We estimate block area by using the published Stratix III data (which is architecturally identical to Stratix IV)
 	      (H. Wong, V. Betz and J. Rose, "Comparing FPGA vs. Custom CMOS and the Impact on Processor Microarchitecture", FPGA 2011) of 0.2623
 		  mm^2 and scaling from 65 to 40 nm to obtain 0.0993 mm^2. That area is for a DSP block with approximately 2x the functionality of
 		  the block we use (can implement two 36x36 multiplies instead of our 1, eight 18x18 multiplies instead of our 4, etc.). Hence we
 		  divide the area by 2 to obtain 0.0497 mm^2. One minimum-width transistor units = 60 L^2 (where L = 40 nm), so is 518,000 MWTUS.
 		  That area includes routing and the connection block input muxes.  Our DSP block is four
 		  rows high, and hence includes four horizontal routing channel segments and four vertical ones, which is 4x the routing of a logic
 		  block (single tile). It also includes 3.6x the outputs of a logic block, and 1.8x the inputs. Hence a slight overestimate of the routing
 		  area associated with our DSP block is four times that of a logic tile, where the routing area of a logic tile was calculated above (at W = 300)
 		  as 30481 MWTAs. Hence the (core, non-routing) area our DSP block is approximately 518,000 - 4 * 30,481 = 396,000 MWTUs.
       -->
     <pb_type name="mult_36">
       <input name="a" num_pins="36"/>
       <input name="b" num_pins="36"/>
       <output name="out" num_pins="72"/>
       <mode name="two_divisible_mult_18x18">
         <pb_type name="divisible_mult_18x18" num_pb="2">
           <input name="a" num_pins="18"/>
           <input name="b" num_pins="18"/>
           <output name="out" num_pins="36"/>
           <!-- Model 9x9 delay and 18x18 delay as the same.  9x9 could be faster, but in Stratix IV
 	          isn't, presumably because the multiplier layout is really optimized for 18x18.
 		-->
           <mode name="two_mult_9x9">
             <pb_type name="mult_9x9_slice" num_pb="2">
               <input name="A_cfg" num_pins="9"/>
               <input name="B_cfg" num_pins="9"/>
               <output name="OUT_cfg" num_pins="18"/>
               <pb_type name="mult_9x9" blif_model=".subckt multiply" num_pb="1">
                 <input name="a" num_pins="9"/>
                 <input name="b" num_pins="9"/>
                 <output name="out" num_pins="18"/>
                 <delay_constant max="1.523e-9" min="0.776e-9" in_port="mult_9x9.a" out_port="mult_9x9.out"/>
                 <delay_constant max="1.523e-9" min="0.776e-9" in_port="mult_9x9.b" out_port="mult_9x9.out"/>
               </pb_type>
               <interconnect>
                 <direct name="a2a" input="mult_9x9_slice.A_cfg" output="mult_9x9.a">
                 </direct>
                 <direct name="b2b" input="mult_9x9_slice.B_cfg" output="mult_9x9.b">
                 </direct>
                 <direct name="out2out" input="mult_9x9.out" output="mult_9x9_slice.OUT_cfg">
                 </direct>
               </interconnect>
               <power method="pin-toggle">
                 <port name="A_cfg" energy_per_toggle="1.45e-12"/>
                 <port name="B_cfg" energy_per_toggle="1.45e-12"/>
                 <static_power power_per_instance="0.0"/>
               </power>
             </pb_type>
             <interconnect>
               <direct name="a2a" input="divisible_mult_18x18.a" output="mult_9x9_slice[1:0].A_cfg">
               </direct>
               <direct name="b2b" input="divisible_mult_18x18.b" output="mult_9x9_slice[1:0].B_cfg">
               </direct>
               <direct name="out2out" input="mult_9x9_slice[1:0].OUT_cfg" output="divisible_mult_18x18.out">
               </direct>
             </interconnect>
           </mode>
           <mode name="mult_18x18">
             <pb_type name="mult_18x18_slice" num_pb="1">
               <input name="A_cfg" num_pins="18"/>
               <input name="B_cfg" num_pins="18"/>
               <output name="OUT_cfg" num_pins="36"/>
               <pb_type name="mult_18x18" blif_model=".subckt multiply" num_pb="1">
                 <input name="a" num_pins="18"/>
                 <input name="b" num_pins="18"/>
                 <output name="out" num_pins="36"/>
                 <delay_constant max="1.523e-9" min="0.776e-9" in_port="mult_18x18.a" out_port="mult_18x18.out"/>
                 <delay_constant max="1.523e-9" min="0.776e-9" in_port="mult_18x18.b" out_port="mult_18x18.out"/>
               </pb_type>
               <interconnect>
                 <direct name="a2a" input="mult_18x18_slice.A_cfg" output="mult_18x18.a">
                 </direct>
                 <direct name="b2b" input="mult_18x18_slice.B_cfg" output="mult_18x18.b">
                 </direct>
                 <direct name="out2out" input="mult_18x18.out" output="mult_18x18_slice.OUT_cfg">
                 </direct>
               </interconnect>
               <power method="pin-toggle">
                 <port name="A_cfg" energy_per_toggle="1.09e-12"/>
                 <port name="B_cfg" energy_per_toggle="1.09e-12"/>
                 <static_power power_per_instance="0.0"/>
               </power>
             </pb_type>
             <interconnect>
               <direct name="a2a" input="divisible_mult_18x18.a" output="mult_18x18_slice.A_cfg">
               </direct>
               <direct name="b2b" input="divisible_mult_18x18.b" output="mult_18x18_slice.B_cfg">
               </direct>
               <direct name="out2out" input="mult_18x18_slice.OUT_cfg" output="divisible_mult_18x18.out">
               </direct>
             </interconnect>
           </mode>
           <power method="sum-of-children"/>
         </pb_type>
         <interconnect>
           <!-- Stratix IV input delay of 207ps is conservative for this architecture because this architecture does not have an input crossbar in the multiplier.
 		   Subtract 72.5 ps delay, which is already in the connection block input mux, leading 134 ps
 				The interconnect difference for DSP blocks is 0.5523, which leads to a minimum delay of 74 ps
               -->
           <direct name="a2a" input="mult_36.a" output="divisible_mult_18x18[1:0].a">
             <delay_constant max="134e-12" min="74e-12" in_port="mult_36.a" out_port="divisible_mult_18x18[1:0].a"/>
           </direct>
           <direct name="b2b" input="mult_36.b" output="divisible_mult_18x18[1:0].b">
             <delay_constant max="134e-12" min="74e-12" in_port="mult_36.b" out_port="divisible_mult_18x18[1:0].b"/>
           </direct>
           <direct name="out2out" input="divisible_mult_18x18[1:0].out" output="mult_36.out">
             <delay_constant max="1.09e-9" min="74e-12" in_port="divisible_mult_18x18[1:0].out" out_port="mult_36.out"/>
           </direct>
         </interconnect>
       </mode>
       <mode name="mult_36x36">
         <pb_type name="mult_36x36_slice" num_pb="1">
           <input name="A_cfg" num_pins="36"/>
           <input name="B_cfg" num_pins="36"/>
           <output name="OUT_cfg" num_pins="72"/>
           <pb_type name="mult_36x36" blif_model=".subckt multiply" num_pb="1">
             <input name="a" num_pins="36"/>
             <input name="b" num_pins="36"/>
             <output name="out" num_pins="72"/>
             <delay_constant max="1.523e-9" min="0.776e-9" in_port="mult_36x36.a" out_port="mult_36x36.out"/>
             <delay_constant max="1.523e-9" min="0.776e-9" in_port="mult_36x36.b" out_port="mult_36x36.out"/>
           </pb_type>
           <interconnect>
             <direct name="a2a" input="mult_36x36_slice.A_cfg" output="mult_36x36.a">
             </direct>
             <direct name="b2b" input="mult_36x36_slice.B_cfg" output="mult_36x36.b">
             </direct>
             <direct name="out2out" input="mult_36x36.out" output="mult_36x36_slice.OUT_cfg">
             </direct>
           </interconnect>
           <power method="pin-toggle">
             <port name="A_cfg" energy_per_toggle="2.13e-12"/>
             <port name="B_cfg" energy_per_toggle="2.13e-12"/>
             <static_power power_per_instance="0.0"/>
           </power>
         </pb_type>
         <interconnect>
           <!-- Stratix IV input delay of 207ps is conservative for this architecture because this architecture does not have an input crossbar in the multiplier.
 		   Subtract 72.5 ps delay, which is already in the connection block input mux, leading
 		   to a 134 ps delay.
 				The interconnect difference for DSP blocks is 0.5523, which leads to a minimum delay of 74 ps
               -->
           <direct name="a2a" input="mult_36.a" output="mult_36x36_slice.A_cfg">
             <delay_constant max="134e-12" min="74e-12" in_port="mult_36.a" out_port="mult_36x36_slice.A_cfg"/>
           </direct>
           <direct name="b2b" input="mult_36.b" output="mult_36x36_slice.B_cfg">
             <delay_constant max="134e-12" min="74e-12" in_port="mult_36.b" out_port="mult_36x36_slice.B_cfg"/>
           </direct>
           <direct name="out2out" input="mult_36x36_slice.OUT_cfg" output="mult_36.out">
             <delay_constant max="1.93e-9" min="74e-12" in_port="mult_36x36_slice.OUT_cfg" out_port="mult_36.out"/>
           </direct>
         </interconnect>
       </mode>
       <!-- Place this multiplier block every 8 columns from (and including) the sixth column -->
       <power method="sum-of-children"/>
     </pb_type>
     <!-- Define fracturable multiplier end -->
     <!-- Define fracturable memory begin -->
     <!-- 32 Kb Memory that can operate from 512x64 to 32Kx1 for single-port mode and 1024x32 to 32Kx1 for dual-port mode.
            Area and max delay based off Stratix IV 9K and 144K memories (delay from linear interpolation, Tsu(483 ps, 636 ps) Tco(1084ps, 1969ps)).

 		   uTh/Tsu ratio = 0.468, uTco min/max ratio = 0.97
            uTh = 226ps, 298ps
            Min uTco = 1051ps, 1909ps
            Max input delay = 204ps (from Stratix IV LAB line) - 72ps (this architecture does not lump connection box delay in internal delay)
            Max output delay = M9K buffer 50ps.
 		   Min input delay = 160ps (from Stratix IV Lab line) - 72ps
 		   Min output delay = 46ps (M9K buffer min/max ratio = 0.9286)

 		   Area is obtained by appropriately scaling and adjusting the published Stratix III (which is architecturally identical to Stratix IV)
 		   data from H. Wong, V. Betz and J. Rose, "Comparing FPGA vs. Custom CMOS and the Impact on Processor Microarchitecture", FPGA 2011.
 		   Linearly interpolating (by bit count) between the M9k and M144k areas to obtain an M32k (our RAM size) point yields a 65 nm area of
 		   of 0.153 mm^2. Interpolating based on port count between the RAMs would instead yield an area of 0.209 mm^2 for our 32 kB RAM; since
 		   bit count accounts for more area than ports for a RAM this size we choose the bit count interpolation; however, since the port interpolation
 		   is not radically different this also gives us confidence that interpolating based on bits is OK, but slightly underpredicts area.
 		   Scaling to 40 nm^2 yields .0579 mm^2, and converting to MWTUs at 60 L^2 / MWTU yields 604,000 MWTUs. This includes routing. A Stratix IV
 		   M9K RAM is one row high and hence has one routing tile (one horizonal and one vertical routing segment area). An M144k RAM has 8 such tiles.
 		   Linearly interpolating on
 		   bits to 32 kb yields 2.2 routing tiles incorporated in the area number above. The inter-block routing represents 30% of the area of a logic
 		   tile according to D. Lewis et al, "Architectural Enhancements in Stratix V," FPGA 2013. Hence we should subtract 0.3 * 2.2 * 84,375 MWTUs to
 		   obtain a RAM core area (not including inter-block routing) of 548,000 MWTU areas for our 32 kb RAM in a 40 nm process.
       -->
     <pb_type name="memory">
       <input name="addr1" num_pins="15"/>
       <input name="addr2" num_pins="15"/>
       <input name="data" num_pins="64"/>
       <input name="we1" num_pins="1"/>
       <input name="we2" num_pins="1"/>
       <output name="out" num_pins="64"/>
       <clock name="clk" num_pins="1"/>
       <!-- Specify single port mode first -->
       <mode name="mem_512x64_sp">
         <pb_type name="mem_512x64_sp" blif_model=".subckt single_port_ram" class="memory" num_pb="1">
           <input name="addr" num_pins="9" port_class="address"/>
           <input name="data" num_pins="64" port_class="data_in"/>
           <input name="we" num_pins="1" port_class="write_en"/>
           <output name="out" num_pins="64" port_class="data_out"/>
           <clock name="clk" num_pins="1" port_class="clock"/>
           <T_setup value="509e-12" port="mem_512x64_sp.addr" clock="clk"/>
           <T_setup value="509e-12" port="mem_512x64_sp.data" clock="clk"/>
           <T_setup value="509e-12" port="mem_512x64_sp.we" clock="clk"/>
           <T_hold value="238e-12" port="mem_512x64_sp.addr" clock="clk"/>
           <T_hold value="238e-12" port="mem_512x64_sp.data" clock="clk"/>
           <T_hold value="238e-12" port="mem_512x64_sp.we" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_512x64_sp.out" clock="clk"/>
           <power method="pin-toggle">
             <port name="clk" energy_per_toggle="9.0e-12"/>
             <static_power power_per_instance="0.0"/>
           </power>
         </pb_type>
         <interconnect>
           <direct name="address1" input="memory.addr1[8:0]" output="mem_512x64_sp.addr">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr1[8:0]" out_port="mem_512x64_sp.addr"/>
           </direct>
           <direct name="data1" input="memory.data[63:0]" output="mem_512x64_sp.data">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[63:0]" out_port="mem_512x64_sp.data"/>
           </direct>
           <direct name="writeen1" input="memory.we1" output="mem_512x64_sp.we">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we1" out_port="mem_512x64_sp.we"/>
           </direct>
           <direct name="dataout1" input="mem_512x64_sp.out" output="memory.out[63:0]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_512x64_sp.out" out_port="memory.out[63:0]"/>
           </direct>
           <direct name="clk" input="memory.clk" output="mem_512x64_sp.clk">
           </direct>
         </interconnect>
       </mode>
       <mode name="mem_1024x32_sp">
         <pb_type name="mem_1024x32_sp" blif_model=".subckt single_port_ram" class="memory" num_pb="1">
           <input name="addr" num_pins="10" port_class="address"/>
           <input name="data" num_pins="32" port_class="data_in"/>
           <input name="we" num_pins="1" port_class="write_en"/>
           <output name="out" num_pins="32" port_class="data_out"/>
           <clock name="clk" num_pins="1" port_class="clock"/>
           <T_setup value="509e-12" port="mem_1024x32_sp.addr" clock="clk"/>
           <T_setup value="509e-12" port="mem_1024x32_sp.data" clock="clk"/>
           <T_setup value="509e-12" port="mem_1024x32_sp.we" clock="clk"/>
           <T_hold value="238e-12" port="mem_1024x32_sp.addr" clock="clk"/>
           <T_hold value="238e-12" port="mem_1024x32_sp.data" clock="clk"/>
           <T_hold value="238e-12" port="mem_1024x32_sp.we" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_1024x32_sp.out" clock="clk"/>
           <power method="pin-toggle">
             <port name="clk" energy_per_toggle="9.0e-12"/>
             <static_power power_per_instance="0.0"/>
           </power>
         </pb_type>
         <interconnect>
           <direct name="address1" input="memory.addr1[9:0]" output="mem_1024x32_sp.addr">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr1[9:0]" out_port="mem_1024x32_sp.addr"/>
           </direct>
           <direct name="data1" input="memory.data[31:0]" output="mem_1024x32_sp.data">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[31:0]" out_port="mem_1024x32_sp.data"/>
           </direct>
           <direct name="writeen1" input="memory.we1" output="mem_1024x32_sp.we">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we1" out_port="mem_1024x32_sp.we"/>
           </direct>
           <direct name="dataout1" input="mem_1024x32_sp.out" output="memory.out[31:0]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_1024x32_sp.out" out_port="memory.out[31:0]"/>
           </direct>
           <direct name="clk" input="memory.clk" output="mem_1024x32_sp.clk">
           </direct>
         </interconnect>
       </mode>
       <mode name="mem_2048x16_sp">
         <pb_type name="mem_2048x16_sp" blif_model=".subckt single_port_ram" class="memory" num_pb="1">
           <input name="addr" num_pins="11" port_class="address"/>
           <input name="data" num_pins="16" port_class="data_in"/>
           <input name="we" num_pins="1" port_class="write_en"/>
           <output name="out" num_pins="16" port_class="data_out"/>
           <clock name="clk" num_pins="1" port_class="clock"/>
           <T_setup value="509e-12" port="mem_2048x16_sp.addr" clock="clk"/>
           <T_setup value="509e-12" port="mem_2048x16_sp.data" clock="clk"/>
           <T_setup value="509e-12" port="mem_2048x16_sp.we" clock="clk"/>
           <T_hold value="238e-12" port="mem_2048x16_sp.addr" clock="clk"/>
           <T_hold value="238e-12" port="mem_2048x16_sp.data" clock="clk"/>
           <T_hold value="238e-12" port="mem_2048x16_sp.we" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_2048x16_sp.out" clock="clk"/>
           <power method="pin-toggle">
             <port name="clk" energy_per_toggle="9.0e-12"/>
             <static_power power_per_instance="0.0"/>
           </power>
         </pb_type>
         <interconnect>
           <direct name="address1" input="memory.addr1[10:0]" output="mem_2048x16_sp.addr">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr1[10:0]" out_port="mem_2048x16_sp.addr"/>
           </direct>
           <direct name="data1" input="memory.data[15:0]" output="mem_2048x16_sp.data">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[15:0]" out_port="mem_2048x16_sp.data"/>
           </direct>
           <direct name="writeen1" input="memory.we1" output="mem_2048x16_sp.we">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we1" out_port="mem_2048x16_sp.we"/>
           </direct>
           <direct name="dataout1" input="mem_2048x16_sp.out" output="memory.out[15:0]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_2048x16_sp.out" out_port="memory.out[15:0]"/>
           </direct>
           <direct name="clk" input="memory.clk" output="mem_2048x16_sp.clk">
           </direct>
         </interconnect>
       </mode>
       <mode name="mem_4096x8_sp">
         <pb_type name="mem_4096x8_sp" blif_model=".subckt single_port_ram" class="memory" num_pb="1">
           <input name="addr" num_pins="12" port_class="address"/>
           <input name="data" num_pins="8" port_class="data_in"/>
           <input name="we" num_pins="1" port_class="write_en"/>
           <output name="out" num_pins="8" port_class="data_out"/>
           <clock name="clk" num_pins="1" port_class="clock"/>
           <T_setup value="509e-12" port="mem_4096x8_sp.addr" clock="clk"/>
           <T_setup value="509e-12" port="mem_4096x8_sp.data" clock="clk"/>
           <T_setup value="509e-12" port="mem_4096x8_sp.we" clock="clk"/>
           <T_hold value="238e-12" port="mem_4096x8_sp.addr" clock="clk"/>
           <T_hold value="238e-12" port="mem_4096x8_sp.data" clock="clk"/>
           <T_hold value="238e-12" port="mem_4096x8_sp.we" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_4096x8_sp.out" clock="clk"/>
           <power method="pin-toggle">
             <port name="clk" energy_per_toggle="9.0e-12"/>
             <static_power power_per_instance="0.0"/>
           </power>
         </pb_type>
         <interconnect>
           <direct name="address1" input="memory.addr1[11:0]" output="mem_4096x8_sp.addr">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr1[11:0]" out_port="mem_4096x8_sp.addr"/>
           </direct>
           <direct name="data1" input="memory.data[7:0]" output="mem_4096x8_sp.data">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[7:0]" out_port="mem_4096x8_sp.data"/>
           </direct>
           <direct name="writeen1" input="memory.we1" output="mem_4096x8_sp.we">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we1" out_port="mem_4096x8_sp.we"/>
           </direct>
           <direct name="dataout1" input="mem_4096x8_sp.out" output="memory.out[7:0]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_4096x8_sp.out" out_port="memory.out[7:0]"/>
           </direct>
           <direct name="clk" input="memory.clk" output="mem_4096x8_sp.clk">
           </direct>
         </interconnect>
       </mode>
       <mode name="mem_8192x4_sp">
         <pb_type name="mem_8192x4_sp" blif_model=".subckt single_port_ram" class="memory" num_pb="1">
           <input name="addr" num_pins="13" port_class="address"/>
           <input name="data" num_pins="4" port_class="data_in"/>
           <input name="we" num_pins="1" port_class="write_en"/>
           <output name="out" num_pins="4" port_class="data_out"/>
           <clock name="clk" num_pins="1" port_class="clock"/>
           <T_setup value="509e-12" port="mem_8192x4_sp.addr" clock="clk"/>
           <T_setup value="509e-12" port="mem_8192x4_sp.data" clock="clk"/>
           <T_setup value="509e-12" port="mem_8192x4_sp.we" clock="clk"/>
           <T_hold value="238e-12" port="mem_8192x4_sp.addr" clock="clk"/>
           <T_hold value="238e-12" port="mem_8192x4_sp.data" clock="clk"/>
           <T_hold value="238e-12" port="mem_8192x4_sp.we" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_8192x4_sp.out" clock="clk"/>
           <power method="pin-toggle">
             <port name="clk" energy_per_toggle="9.0e-12"/>
             <static_power power_per_instance="0.0"/>
           </power>
         </pb_type>
         <interconnect>
           <direct name="address1" input="memory.addr1[12:0]" output="mem_8192x4_sp.addr">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr1[12:0]" out_port="mem_8192x4_sp.addr"/>
           </direct>
           <direct name="data1" input="memory.data[3:0]" output="mem_8192x4_sp.data">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[3:0]" out_port="mem_8192x4_sp.data"/>
           </direct>
           <direct name="writeen1" input="memory.we1" output="mem_8192x4_sp.we">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we1" out_port="mem_8192x4_sp.we"/>
           </direct>
           <direct name="dataout1" input="mem_8192x4_sp.out" output="memory.out[3:0]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_8192x4_sp.out" out_port="memory.out[3:0]"/>
           </direct>
           <direct name="clk" input="memory.clk" output="mem_8192x4_sp.clk">
           </direct>
         </interconnect>
       </mode>
       <mode name="mem_16384x2_sp">
         <pb_type name="mem_16384x2_sp" blif_model=".subckt single_port_ram" class="memory" num_pb="1">
           <input name="addr" num_pins="14" port_class="address"/>
           <input name="data" num_pins="2" port_class="data_in"/>
           <input name="we" num_pins="1" port_class="write_en"/>
           <output name="out" num_pins="2" port_class="data_out"/>
           <clock name="clk" num_pins="1" port_class="clock"/>
           <T_setup value="509e-12" port="mem_16384x2_sp.addr" clock="clk"/>
           <T_setup value="509e-12" port="mem_16384x2_sp.data" clock="clk"/>
           <T_setup value="509e-12" port="mem_16384x2_sp.we" clock="clk"/>
           <T_hold value="238e-12" port="mem_16384x2_sp.addr" clock="clk"/>
           <T_hold value="238e-12" port="mem_16384x2_sp.data" clock="clk"/>
           <T_hold value="238e-12" port="mem_16384x2_sp.we" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_16384x2_sp.out" clock="clk"/>
           <power method="pin-toggle">
             <port name="clk" energy_per_toggle="9.0e-12"/>
             <static_power power_per_instance="0.0"/>
           </power>
         </pb_type>
         <interconnect>
           <direct name="address1" input="memory.addr1[13:0]" output="mem_16384x2_sp.addr">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr1[13:0]" out_port="mem_16384x2_sp.addr"/>
           </direct>
           <direct name="data1" input="memory.data[1:0]" output="mem_16384x2_sp.data">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[1:0]" out_port="mem_16384x2_sp.data"/>
           </direct>
           <direct name="writeen1" input="memory.we1" output="mem_16384x2_sp.we">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we1" out_port="mem_16384x2_sp.we"/>
           </direct>
           <direct name="dataout1" input="mem_16384x2_sp.out" output="memory.out[1:0]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_16384x2_sp.out" out_port="memory.out[1:0]"/>
           </direct>
           <direct name="clk" input="memory.clk" output="mem_16384x2_sp.clk">
           </direct>
         </interconnect>
       </mode>
       <mode name="mem_32768x1_sp">
         <pb_type name="mem_32768x1_sp" blif_model=".subckt single_port_ram" class="memory" num_pb="1">
           <input name="addr" num_pins="15" port_class="address"/>
           <input name="data" num_pins="1" port_class="data_in"/>
           <input name="we" num_pins="1" port_class="write_en"/>
           <output name="out" num_pins="1" port_class="data_out"/>
           <clock name="clk" num_pins="1" port_class="clock"/>
           <T_setup value="509e-12" port="mem_32768x1_sp.addr" clock="clk"/>
           <T_setup value="509e-12" port="mem_32768x1_sp.data" clock="clk"/>
           <T_setup value="509e-12" port="mem_32768x1_sp.we" clock="clk"/>
           <T_hold value="238e-12" port="mem_32768x1_sp.addr" clock="clk"/>
           <T_hold value="238e-12" port="mem_32768x1_sp.data" clock="clk"/>
           <T_hold value="238e-12" port="mem_32768x1_sp.we" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_32768x1_sp.out" clock="clk"/>
           <power method="pin-toggle">
             <port name="clk" energy_per_toggle="9.0e-12"/>
             <static_power power_per_instance="0.0"/>
           </power>
         </pb_type>
         <interconnect>
           <direct name="address1" input="memory.addr1[14:0]" output="mem_32768x1_sp.addr">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr1[14:0]" out_port="mem_32768x1_sp.addr"/>
           </direct>
           <direct name="data1" input="memory.data[0:0]" output="mem_32768x1_sp.data">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[0:0]" out_port="mem_32768x1_sp.data"/>
           </direct>
           <direct name="writeen1" input="memory.we1" output="mem_32768x1_sp.we">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we1" out_port="mem_32768x1_sp.we"/>
           </direct>
           <direct name="dataout1" input="mem_32768x1_sp.out" output="memory.out[0:0]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_32768x1_sp.out" out_port="memory.out[0:0]"/>
           </direct>
           <direct name="clk" input="memory.clk" output="mem_32768x1_sp.clk">
           </direct>
         </interconnect>
       </mode>
       <!-- Specify true dual port mode next -->
       <mode name="mem_1024x32_dp">
         <pb_type name="mem_1024x32_dp" blif_model=".subckt dual_port_ram" class="memory" num_pb="1">
           <input name="addr1" num_pins="10" port_class="address1"/>
           <input name="addr2" num_pins="10" port_class="address2"/>
           <input name="data1" num_pins="32" port_class="data_in1"/>
           <input name="data2" num_pins="32" port_class="data_in2"/>
           <input name="we1" num_pins="1" port_class="write_en1"/>
           <input name="we2" num_pins="1" port_class="write_en2"/>
           <output name="out1" num_pins="32" port_class="data_out1"/>
           <output name="out2" num_pins="32" port_class="data_out2"/>
           <clock name="clk" num_pins="1" port_class="clock"/>
           <T_setup value="509e-12" port="mem_1024x32_dp.addr1" clock="clk"/>
           <T_setup value="509e-12" port="mem_1024x32_dp.data1" clock="clk"/>
           <T_setup value="509e-12" port="mem_1024x32_dp.we1" clock="clk"/>
           <T_setup value="509e-12" port="mem_1024x32_dp.addr2" clock="clk"/>
           <T_setup value="509e-12" port="mem_1024x32_dp.data2" clock="clk"/>
           <T_setup value="509e-12" port="mem_1024x32_dp.we2" clock="clk"/>
           <T_hold value="238e-12" port="mem_1024x32_dp.addr1" clock="clk"/>
           <T_hold value="238e-12" port="mem_1024x32_dp.data1" clock="clk"/>
           <T_hold value="238e-12" port="mem_1024x32_dp.we1" clock="clk"/>
           <T_hold value="238e-12" port="mem_1024x32_dp.addr2" clock="clk"/>
           <T_hold value="238e-12" port="mem_1024x32_dp.data2" clock="clk"/>
           <T_hold value="238e-12" port="mem_1024x32_dp.we2" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_1024x32_dp.out1" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_1024x32_dp.out2" clock="clk"/>
           <power method="pin-toggle">
             <port name="clk" energy_per_toggle="17.9e-12"/>
             <static_power power_per_instance="0.0"/>
           </power>
         </pb_type>
         <interconnect>
           <direct name="address1" input="memory.addr1[9:0]" output="mem_1024x32_dp.addr1">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr1[9:0]" out_port="mem_1024x32_dp.addr1"/>
           </direct>
           <direct name="address2" input="memory.addr2[9:0]" output="mem_1024x32_dp.addr2">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr2[9:0]" out_port="mem_1024x32_dp.addr2"/>
           </direct>
           <direct name="data1" input="memory.data[31:0]" output="mem_1024x32_dp.data1">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[31:0]" out_port="mem_1024x32_dp.data1"/>
           </direct>
           <direct name="data2" input="memory.data[63:32]" output="mem_1024x32_dp.data2">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[63:32]" out_port="mem_1024x32_dp.data2"/>
           </direct>
           <direct name="writeen1" input="memory.we1" output="mem_1024x32_dp.we1">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we1" out_port="mem_1024x32_dp.we1"/>
           </direct>
           <direct name="writeen2" input="memory.we2" output="mem_1024x32_dp.we2">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we2" out_port="mem_1024x32_dp.we2"/>
           </direct>
           <direct name="dataout1" input="mem_1024x32_dp.out1" output="memory.out[31:0]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_1024x32_dp.out1" out_port="memory.out[31:0]"/>
           </direct>
           <direct name="dataout2" input="mem_1024x32_dp.out2" output="memory.out[63:32]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_1024x32_dp.out2" out_port="memory.out[63:32]"/>
           </direct>
           <direct name="clk" input="memory.clk" output="mem_1024x32_dp.clk">
           </direct>
         </interconnect>
       </mode>
       <mode name="mem_2048x16_dp">
         <pb_type name="mem_2048x16_dp" blif_model=".subckt dual_port_ram" class="memory" num_pb="1">
           <input name="addr1" num_pins="11" port_class="address1"/>
           <input name="addr2" num_pins="11" port_class="address2"/>
           <input name="data1" num_pins="16" port_class="data_in1"/>
           <input name="data2" num_pins="16" port_class="data_in2"/>
           <input name="we1" num_pins="1" port_class="write_en1"/>
           <input name="we2" num_pins="1" port_class="write_en2"/>
           <output name="out1" num_pins="16" port_class="data_out1"/>
           <output name="out2" num_pins="16" port_class="data_out2"/>
           <clock name="clk" num_pins="1" port_class="clock"/>
           <T_setup value="509e-12" port="mem_2048x16_dp.addr1" clock="clk"/>
           <T_setup value="509e-12" port="mem_2048x16_dp.data1" clock="clk"/>
           <T_setup value="509e-12" port="mem_2048x16_dp.we1" clock="clk"/>
           <T_setup value="509e-12" port="mem_2048x16_dp.addr2" clock="clk"/>
           <T_setup value="509e-12" port="mem_2048x16_dp.data2" clock="clk"/>
           <T_setup value="509e-12" port="mem_2048x16_dp.we2" clock="clk"/>
           <T_hold value="238e-12" port="mem_2048x16_dp.addr1" clock="clk"/>
           <T_hold value="238e-12" port="mem_2048x16_dp.data1" clock="clk"/>
           <T_hold value="238e-12" port="mem_2048x16_dp.we1" clock="clk"/>
           <T_hold value="238e-12" port="mem_2048x16_dp.addr2" clock="clk"/>
           <T_hold value="238e-12" port="mem_2048x16_dp.data2" clock="clk"/>
           <T_hold value="238e-12" port="mem_2048x16_dp.we2" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_2048x16_dp.out1" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_2048x16_dp.out2" clock="clk"/>
           <power method="pin-toggle">
             <port name="clk" energy_per_toggle="17.9e-12"/>
             <static_power power_per_instance="0.0"/>
           </power>
         </pb_type>
         <interconnect>
           <direct name="address1" input="memory.addr1[10:0]" output="mem_2048x16_dp.addr1">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr1[10:0]" out_port="mem_2048x16_dp.addr1"/>
           </direct>
           <direct name="address2" input="memory.addr2[10:0]" output="mem_2048x16_dp.addr2">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr2[10:0]" out_port="mem_2048x16_dp.addr2"/>
           </direct>
           <direct name="data1" input="memory.data[15:0]" output="mem_2048x16_dp.data1">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[15:0]" out_port="mem_2048x16_dp.data1"/>
           </direct>
           <direct name="data2" input="memory.data[31:16]" output="mem_2048x16_dp.data2">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[31:16]" out_port="mem_2048x16_dp.data2"/>
           </direct>
           <direct name="writeen1" input="memory.we1" output="mem_2048x16_dp.we1">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we1" out_port="mem_2048x16_dp.we1"/>
           </direct>
           <direct name="writeen2" input="memory.we2" output="mem_2048x16_dp.we2">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we2" out_port="mem_2048x16_dp.we2"/>
           </direct>
           <direct name="dataout1" input="mem_2048x16_dp.out1" output="memory.out[15:0]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_2048x16_dp.out1" out_port="memory.out[15:0]"/>
           </direct>
           <direct name="dataout2" input="mem_2048x16_dp.out2" output="memory.out[31:16]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_2048x16_dp.out2" out_port="memory.out[31:16]"/>
           </direct>
           <direct name="clk" input="memory.clk" output="mem_2048x16_dp.clk">
           </direct>
         </interconnect>
       </mode>
       <mode name="mem_2048x8_dp">
         <pb_type name="mem_2048x8_dp" blif_model=".subckt dual_port_ram" class="memory" num_pb="1">
           <input name="addr1" num_pins="12" port_class="address1"/>
           <input name="addr2" num_pins="12" port_class="address2"/>
           <input name="data1" num_pins="8" port_class="data_in1"/>
           <input name="data2" num_pins="8" port_class="data_in2"/>
           <input name="we1" num_pins="1" port_class="write_en1"/>
           <input name="we2" num_pins="1" port_class="write_en2"/>
           <output name="out1" num_pins="8" port_class="data_out1"/>
           <output name="out2" num_pins="8" port_class="data_out2"/>
           <clock name="clk" num_pins="1" port_class="clock"/>
           <T_setup value="509e-12" port="mem_2048x8_dp.addr1" clock="clk"/>
           <T_setup value="509e-12" port="mem_2048x8_dp.data1" clock="clk"/>
           <T_setup value="509e-12" port="mem_2048x8_dp.we1" clock="clk"/>
           <T_setup value="509e-12" port="mem_2048x8_dp.addr2" clock="clk"/>
           <T_setup value="509e-12" port="mem_2048x8_dp.data2" clock="clk"/>
           <T_setup value="509e-12" port="mem_2048x8_dp.we2" clock="clk"/>
           <T_hold value="238e-12" port="mem_2048x8_dp.addr1" clock="clk"/>
           <T_hold value="238e-12" port="mem_2048x8_dp.data1" clock="clk"/>
           <T_hold value="238e-12" port="mem_2048x8_dp.we1" clock="clk"/>
           <T_hold value="238e-12" port="mem_2048x8_dp.addr2" clock="clk"/>
           <T_hold value="238e-12" port="mem_2048x8_dp.data2" clock="clk"/>
           <T_hold value="238e-12" port="mem_2048x8_dp.we2" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_2048x8_dp.out1" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_2048x8_dp.out2" clock="clk"/>
           <power method="pin-toggle">
             <port name="clk" energy_per_toggle="17.9e-12"/>
             <static_power power_per_instance="0.0"/>
           </power>
         </pb_type>
         <interconnect>
           <direct name="address1" input="memory.addr1[11:0]" output="mem_2048x8_dp.addr1">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr1[11:0]" out_port="mem_2048x8_dp.addr1"/>
           </direct>
           <direct name="address2" input="memory.addr2[11:0]" output="mem_2048x8_dp.addr2">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr2[11:0]" out_port="mem_2048x8_dp.addr2"/>
           </direct>
           <direct name="data1" input="memory.data[7:0]" output="mem_2048x8_dp.data1">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[7:0]" out_port="mem_2048x8_dp.data1"/>
           </direct>
           <direct name="data2" input="memory.data[15:8]" output="mem_2048x8_dp.data2">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[15:8]" out_port="mem_2048x8_dp.data2"/>
           </direct>
           <direct name="writeen1" input="memory.we1" output="mem_2048x8_dp.we1">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we1" out_port="mem_2048x8_dp.we1"/>
           </direct>
           <direct name="writeen2" input="memory.we2" output="mem_2048x8_dp.we2">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we2" out_port="mem_2048x8_dp.we2"/>
           </direct>
           <direct name="dataout1" input="mem_2048x8_dp.out1" output="memory.out[7:0]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_2048x8_dp.out1" out_port="memory.out[7:0]"/>
           </direct>
           <direct name="dataout2" input="mem_2048x8_dp.out2" output="memory.out[15:8]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_2048x8_dp.out2" out_port="memory.out[15:8]"/>
           </direct>
           <direct name="clk" input="memory.clk" output="mem_2048x8_dp.clk">
           </direct>
         </interconnect>
       </mode>
       <mode name="mem_8192x4_dp">
         <pb_type name="mem_8192x4_dp" blif_model=".subckt dual_port_ram" class="memory" num_pb="1">
           <input name="addr1" num_pins="13" port_class="address1"/>
           <input name="addr2" num_pins="13" port_class="address2"/>
           <input name="data1" num_pins="4" port_class="data_in1"/>
           <input name="data2" num_pins="4" port_class="data_in2"/>
           <input name="we1" num_pins="1" port_class="write_en1"/>
           <input name="we2" num_pins="1" port_class="write_en2"/>
           <output name="out1" num_pins="4" port_class="data_out1"/>
           <output name="out2" num_pins="4" port_class="data_out2"/>
           <clock name="clk" num_pins="1" port_class="clock"/>
           <T_setup value="509e-12" port="mem_8192x4_dp.addr1" clock="clk"/>
           <T_setup value="509e-12" port="mem_8192x4_dp.data1" clock="clk"/>
           <T_setup value="509e-12" port="mem_8192x4_dp.we1" clock="clk"/>
           <T_setup value="509e-12" port="mem_8192x4_dp.addr2" clock="clk"/>
           <T_setup value="509e-12" port="mem_8192x4_dp.data2" clock="clk"/>
           <T_setup value="509e-12" port="mem_8192x4_dp.we2" clock="clk"/>
           <T_hold value="238e-12" port="mem_8192x4_dp.addr1" clock="clk"/>
           <T_hold value="238e-12" port="mem_8192x4_dp.data1" clock="clk"/>
           <T_hold value="238e-12" port="mem_8192x4_dp.we1" clock="clk"/>
           <T_hold value="238e-12" port="mem_8192x4_dp.addr2" clock="clk"/>
           <T_hold value="238e-12" port="mem_8192x4_dp.data2" clock="clk"/>
           <T_hold value="238e-12" port="mem_8192x4_dp.we2" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_8192x4_dp.out1" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_8192x4_dp.out2" clock="clk"/>
           <power method="pin-toggle">
             <port name="clk" energy_per_toggle="17.9e-12"/>
             <static_power power_per_instance="0.0"/>
           </power>
         </pb_type>
         <interconnect>
           <direct name="address1" input="memory.addr1[12:0]" output="mem_8192x4_dp.addr1">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr1[12:0]" out_port="mem_8192x4_dp.addr1"/>
           </direct>
           <direct name="address2" input="memory.addr2[12:0]" output="mem_8192x4_dp.addr2">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr2[12:0]" out_port="mem_8192x4_dp.addr2"/>
           </direct>
           <direct name="data1" input="memory.data[3:0]" output="mem_8192x4_dp.data1">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[3:0]" out_port="mem_8192x4_dp.data1"/>
           </direct>
           <direct name="data2" input="memory.data[7:4]" output="mem_8192x4_dp.data2">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[7:4]" out_port="mem_8192x4_dp.data2"/>
           </direct>
           <direct name="writeen1" input="memory.we1" output="mem_8192x4_dp.we1">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we1" out_port="mem_8192x4_dp.we1"/>
           </direct>
           <direct name="writeen2" input="memory.we2" output="mem_8192x4_dp.we2">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we2" out_port="mem_8192x4_dp.we2"/>
           </direct>
           <direct name="dataout1" input="mem_8192x4_dp.out1" output="memory.out[3:0]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_8192x4_dp.out1" out_port="memory.out[3:0]"/>
           </direct>
           <direct name="dataout2" input="mem_8192x4_dp.out2" output="memory.out[7:4]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_8192x4_dp.out2" out_port="memory.out[7:4]"/>
           </direct>
           <direct name="clk" input="memory.clk" output="mem_8192x4_dp.clk">
           </direct>
         </interconnect>
       </mode>
       <mode name="mem_16384x2_dp">
         <pb_type name="mem_16384x2_dp" blif_model=".subckt dual_port_ram" class="memory" num_pb="1">
           <input name="addr1" num_pins="14" port_class="address1"/>
           <input name="addr2" num_pins="14" port_class="address2"/>
           <input name="data1" num_pins="2" port_class="data_in1"/>
           <input name="data2" num_pins="2" port_class="data_in2"/>
           <input name="we1" num_pins="1" port_class="write_en1"/>
           <input name="we2" num_pins="1" port_class="write_en2"/>
           <output name="out1" num_pins="2" port_class="data_out1"/>
           <output name="out2" num_pins="2" port_class="data_out2"/>
           <clock name="clk" num_pins="1" port_class="clock"/>
           <T_setup value="509e-12" port="mem_16384x2_dp.addr1" clock="clk"/>
           <T_setup value="509e-12" port="mem_16384x2_dp.data1" clock="clk"/>
           <T_setup value="509e-12" port="mem_16384x2_dp.we1" clock="clk"/>
           <T_setup value="509e-12" port="mem_16384x2_dp.addr2" clock="clk"/>
           <T_setup value="509e-12" port="mem_16384x2_dp.data2" clock="clk"/>
           <T_setup value="509e-12" port="mem_16384x2_dp.we2" clock="clk"/>
           <T_hold value="238e-12" port="mem_16384x2_dp.addr1" clock="clk"/>
           <T_hold value="238e-12" port="mem_16384x2_dp.data1" clock="clk"/>
           <T_hold value="238e-12" port="mem_16384x2_dp.we1" clock="clk"/>
           <T_hold value="238e-12" port="mem_16384x2_dp.addr2" clock="clk"/>
           <T_hold value="238e-12" port="mem_16384x2_dp.data2" clock="clk"/>
           <T_hold value="238e-12" port="mem_16384x2_dp.we2" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_16384x2_dp.out1" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_16384x2_dp.out2" clock="clk"/>
           <power method="pin-toggle">
             <port name="clk" energy_per_toggle="17.9e-12"/>
             <static_power power_per_instance="0.0"/>
           </power>
         </pb_type>
         <interconnect>
           <direct name="address1" input="memory.addr1[13:0]" output="mem_16384x2_dp.addr1">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr1[13:0]" out_port="mem_16384x2_dp.addr1"/>
           </direct>
           <direct name="address2" input="memory.addr2[13:0]" output="mem_16384x2_dp.addr2">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr2[13:0]" out_port="mem_16384x2_dp.addr2"/>
           </direct>
           <direct name="data1" input="memory.data[1:0]" output="mem_16384x2_dp.data1">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[1:0]" out_port="mem_16384x2_dp.data1"/>
           </direct>
           <direct name="data2" input="memory.data[3:2]" output="mem_16384x2_dp.data2">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[3:2]" out_port="mem_16384x2_dp.data2"/>
           </direct>
           <direct name="writeen1" input="memory.we1" output="mem_16384x2_dp.we1">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we1" out_port="mem_16384x2_dp.we1"/>
           </direct>
           <direct name="writeen2" input="memory.we2" output="mem_16384x2_dp.we2">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we2" out_port="mem_16384x2_dp.we2"/>
           </direct>
           <direct name="dataout1" input="mem_16384x2_dp.out1" output="memory.out[1:0]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_16384x2_dp.out1" out_port="memory.out[1:0]"/>
           </direct>
           <direct name="dataout2" input="mem_16384x2_dp.out2" output="memory.out[3:2]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_16384x2_dp.out2" out_port="memory.out[3:2]"/>
           </direct>
           <direct name="clk" input="memory.clk" output="mem_16384x2_dp.clk">
           </direct>
         </interconnect>
       </mode>
       <mode name="mem_32768x1_dp">
         <pb_type name="mem_32768x1_dp" blif_model=".subckt dual_port_ram" class="memory" num_pb="1">
           <input name="addr1" num_pins="15" port_class="address1"/>
           <input name="addr2" num_pins="15" port_class="address2"/>
           <input name="data1" num_pins="1" port_class="data_in1"/>
           <input name="data2" num_pins="1" port_class="data_in2"/>
           <input name="we1" num_pins="1" port_class="write_en1"/>
           <input name="we2" num_pins="1" port_class="write_en2"/>
           <output name="out1" num_pins="1" port_class="data_out1"/>
           <output name="out2" num_pins="1" port_class="data_out2"/>
           <clock name="clk" num_pins="1" port_class="clock"/>
           <T_setup value="509e-12" port="mem_32768x1_dp.addr1" clock="clk"/>
           <T_setup value="509e-12" port="mem_32768x1_dp.data1" clock="clk"/>
           <T_setup value="509e-12" port="mem_32768x1_dp.we1" clock="clk"/>
           <T_setup value="509e-12" port="mem_32768x1_dp.addr2" clock="clk"/>
           <T_setup value="509e-12" port="mem_32768x1_dp.data2" clock="clk"/>
           <T_setup value="509e-12" port="mem_32768x1_dp.we2" clock="clk"/>
           <T_hold value="238e-12" port="mem_32768x1_dp.addr1" clock="clk"/>
           <T_hold value="238e-12" port="mem_32768x1_dp.data1" clock="clk"/>
           <T_hold value="238e-12" port="mem_32768x1_dp.we1" clock="clk"/>
           <T_hold value="238e-12" port="mem_32768x1_dp.addr2" clock="clk"/>
           <T_hold value="238e-12" port="mem_32768x1_dp.data2" clock="clk"/>
           <T_hold value="238e-12" port="mem_32768x1_dp.we2" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_32768x1_dp.out1" clock="clk"/>
           <T_clock_to_Q max="1.234e-9" min="1.196e-9" port="mem_32768x1_dp.out2" clock="clk"/>
           <power method="pin-toggle">
             <port name="clk" energy_per_toggle="17.9e-12"/>
             <static_power power_per_instance="0.0"/>
           </power>
         </pb_type>
         <interconnect>
           <direct name="address1" input="memory.addr1[14:0]" output="mem_32768x1_dp.addr1">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr1[14:0]" out_port="mem_32768x1_dp.addr1"/>
           </direct>
           <direct name="address2" input="memory.addr2[14:0]" output="mem_32768x1_dp.addr2">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.addr2[14:0]" out_port="mem_32768x1_dp.addr2"/>
           </direct>
           <direct name="data1" input="memory.data[0:0]" output="mem_32768x1_dp.data1">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[0:0]" out_port="mem_32768x1_dp.data1"/>
           </direct>
           <direct name="data2" input="memory.data[1:1]" output="mem_32768x1_dp.data2">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.data[1:1]" out_port="mem_32768x1_dp.data2"/>
           </direct>
           <direct name="writeen1" input="memory.we1" output="mem_32768x1_dp.we1">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we1" out_port="mem_32768x1_dp.we1"/>
           </direct>
           <direct name="writeen2" input="memory.we2" output="mem_32768x1_dp.we2">
             <delay_constant max="132e-12" min="88e-12" in_port="memory.we2" out_port="mem_32768x1_dp.we2"/>
           </direct>
           <direct name="dataout1" input="mem_32768x1_dp.out1" output="memory.out[0:0]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_32768x1_dp.out1" out_port="memory.out[0:0]"/>
           </direct>
           <direct name="dataout2" input="mem_32768x1_dp.out2" output="memory.out[1:1]">
             <delay_constant max="50e-12" min="46e-12" in_port="mem_32768x1_dp.out2" out_port="memory.out[1:1]"/>
           </direct>
           <direct name="clk" input="memory.clk" output="mem_32768x1_dp.clk">
           </direct>
         </interconnect>
       </mode>
       <!-- Every input pin is driven by 15% of the tracks in a channel, every output pin is driven by 10% of the tracks in a channel -->
       <!-- Place this memory block every 8 columns from (and including) the second column -->
       <power method="sum-of-children"/>
     </pb_type>
     <!-- Define fracturable memory end -->
   </complexblocklist>
   <power>
     <local_interconnect C_wire="2.5e-10"/>
     <mux_transistor_size mux_transistor_size="3"/>
     <FF_size FF_size="4"/>
     <LUT_transistor_size LUT_transistor_size="4"/>
   </power>
   <clocks>
     <clock buffer_size="auto" C_wire="2.5e-10"/>
   </clocks>
 </architecture>