[Arch] Update timing for SOFA architecture

This commit is contained in:
tangxifan 2021-04-01 16:39:19 -06:00
parent 1b59daebc6
commit 062120ffd9
1 changed files with 75 additions and 114 deletions

View File

@ -11,6 +11,9 @@
- 80% L = 4, fc_in = 0.15, Fc_out = 0.10
- 100 routing tracks per channel
- The timing is extracted from a TT corner (1.8V, 25C)
- TODO: Add multi-corners, in particular, SS corner
Authors: Xifan Tang
-->
<architecture>
@ -186,21 +189,6 @@
</fixed_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.
@ -214,41 +202,32 @@
<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="L1_mux" R="551" Cin=".77e-15" Cout="4e-15" Tdel="58e-12" mux_trans_size="2.630740" buf_size="27.645901"/>
<switch type="mux" name="L2_mux" R="551" Cin=".77e-15" Cout="4e-15" Tdel="58e-12" mux_trans_size="2.630740" buf_size="27.645901"/>
<switch type="mux" name="L4_mux" R="551" Cin=".77e-15" Cout="4e-15" Tdel="58e-12" mux_trans_size="2.630740" buf_size="27.645901"/>
<!-- Give uniform delays for all the MUXes driving different length of wires
TODO: Can be more accurate once the report timing strategies are elaborated
-->
<switch type="mux" name="L1_mux" R="0" Cin="0" Cout="0" Tdel="1.61e-9" mux_trans_size="2.630740" buf_size="27.645901"/>
<switch type="mux" name="L2_mux" R="0" Cin="0" Cout="0" Tdel="1.61e-9" mux_trans_size="2.630740" buf_size="27.645901"/>
<switch type="mux" name="L4_mux" R="0" Cin="0" Cout="0" Tdel="1.61e-9" 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"/>
<switch type="mux" name="ipin_cblock" R="0" Cout="0." Cin="0" Tdel="1.38e-9" 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. -->
<!--- The wire delay is around 0.1ns in post PnR netlist.
Create a pair of RC value so that R * C = 0.1ns
This is o.k. because other RC values are all zero
-->
<!-- GIVE a specific name for the segment! OpenFPGA appreciate that! -->
<segment name="L1" freq="0.10" length="1" type="unidir" Rmetal="101" Cmetal="22.5e-15">
<segment name="L1" freq="0.10" length="1" type="unidir" Rmetal="100" Cmetal="1e-12">
<mux name="L1_mux"/>
<sb type="pattern">1 1</sb>
<cb type="pattern">1</cb>
</segment>
<segment name="L2" freq="0.10" length="2" type="unidir" Rmetal="101" Cmetal="22.5e-15">
<segment name="L2" freq="0.10" length="2" type="unidir" Rmetal="100" Cmetal="1e-12">
<mux name="L2_mux"/>
<sb type="pattern">1 1 1</sb>
<cb type="pattern">1 1</cb>
</segment>
<segment name="L4" freq="0.80" length="4" type="unidir" Rmetal="101" Cmetal="22.5e-15">
<segment name="L4" freq="0.80" length="4" type="unidir" Rmetal="100" Cmetal="1e-12">
<mux name="L4_mux"/>
<sb type="pattern">1 1 1 1 1</sb>
<cb type="pattern">1 1 1 1</cb>
@ -277,18 +256,17 @@
</pb_type>
<interconnect>
<direct name="outpad" input="io.outpad" output="iopad.outpad">
<delay_constant max="1.394e-11" in_port="io.outpad" out_port="iopad.outpad"/>
<delay_constant max="0.11e-9" in_port="io.outpad" out_port="iopad.outpad"/>
</direct>
<direct name="inpad" input="iopad.inpad" output="io.inpad">
<delay_constant max="4.243e-11" in_port="iopad.inpad" out_port="io.inpad"/>
<delay_constant max="0.11e-9" in_port="iopad.inpad" out_port="io.inpad"/>
</direct>
</interconnect>
</mode>
<!-- IOs can operate as either inputs or outputs.
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.
The Embedded I/O timing is 0.11ns
FIXME: the timing may include the GPIO timing!!!
-->
<mode name="inpad">
<pb_type name="inpad" blif_model=".input" num_pb="1">
@ -296,7 +274,7 @@
</pb_type>
<interconnect>
<direct name="inpad" input="inpad.inpad" output="io.inpad">
<delay_constant max="4.243e-11" in_port="inpad.inpad" out_port="io.inpad"/>
<delay_constant max="0.11e-9" in_port="inpad.inpad" out_port="io.inpad"/>
</direct>
</interconnect>
</mode>
@ -306,7 +284,7 @@
</pb_type>
<interconnect>
<direct name="outpad" input="io.outpad" output="outpad.outpad">
<delay_constant max="1.394e-11" in_port="io.outpad" out_port="outpad.outpad"/>
<delay_constant max="0.11e-9" in_port="io.outpad" out_port="outpad.outpad"/>
</direct>
</interconnect>
</mode>
@ -386,9 +364,9 @@
<input name="DI" num_pins="1"/>
<output name="Q" num_pins="1"/>
<clock name="clk" num_pins="1"/>
<T_setup value="66e-12" port="ff.D" clock="clk"/>
<T_setup value="66e-12" port="ff.DI" clock="clk"/>
<T_clock_to_Q max="124e-12" port="ff.Q" clock="clk"/>
<T_setup value="0.39e-9" port="ff.D" clock="clk"/>
<T_setup value="0.39e-9" port="ff.DI" clock="clk"/>
<T_clock_to_Q max="0.43e-9" port="ff.Q" clock="clk"/>
</pb_type>
<interconnect>
<direct name="direct1" input="fabric.in" output="frac_logic.in"/>
@ -398,22 +376,22 @@
<direct name="direct5" input="ff[1].Q" output="fabric.reg_out"/>
<complete name="complete1" input="fabric.clk" output="ff[1:0].clk"/>
<mux name="mux1" input="frac_logic.out[0:0] fabric.reg_in" output="ff[0:0].D">
<delay_constant max="25e-12" in_port="frac_logic.out[0:0]" out_port="ff[0:0].D"/>
<delay_constant max="45e-12" in_port="fabric.reg_in" out_port="ff[0:0].D"/>
<delay_constant max="1.14e-9" in_port="frac_logic.out[0:0]" out_port="ff[0:0].D"/>
<delay_constant max="1.14e-9" in_port="fabric.reg_in" out_port="ff[0:0].D"/>
</mux>
<mux name="mux2" input="frac_logic.out[1:1] ff[0:0].Q" output="ff[1:1].D">
<delay_constant max="25e-12" in_port="frac_logic.out[1:1]" out_port="ff[1:1].D"/>
<delay_constant max="45e-12" in_port="ff[0:0].Q" out_port="ff[1:1].D"/>
<delay_constant max="0.56e-9" in_port="frac_logic.out[1:1]" out_port="ff[1:1].D"/>
<delay_constant max="0.56e-9" in_port="ff[0:0].Q" out_port="ff[1:1].D"/>
</mux>
<mux name="mux3" input="ff[0].Q frac_logic.out[0]" output="fabric.out[0]">
<!-- LUT to output is faster than FF to output on a Stratix IV -->
<delay_constant max="25e-12" in_port="frac_logic.out[0]" out_port="fabric.out[0]"/>
<delay_constant max="45e-12" in_port="ff[0].Q" out_port="fabric.out[0]"/>
<delay_constant max="0.89e-9" in_port="frac_logic.out[0]" out_port="fabric.out[0]"/>
<delay_constant max="0.88e-9" in_port="ff[0].Q" out_port="fabric.out[0]"/>
</mux>
<mux name="mux4" input="ff[1].Q frac_logic.out[1]" output="fabric.out[1]">
<!-- LUT to output is faster than FF to output on a Stratix IV -->
<delay_constant max="25e-12" in_port="frac_logic.out[1]" out_port="fabric.out[1]"/>
<delay_constant max="45e-12" in_port="ff[1].Q" out_port="fabric.out[1]"/>
<delay_constant max="0.78e-9" in_port="frac_logic.out[1]" out_port="fabric.out[1]"/>
<delay_constant max="0.89e-9" in_port="ff[1].Q" out_port="fabric.out[1]"/>
</mux>
</interconnect>
</pb_type>
@ -443,18 +421,10 @@
<input name="in" num_pins="3" 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
-->
<delay_matrix type="max" in_port="lut3.in" out_port="lut3.out">
235e-12
235e-12
235e-12
2.31e-9
2.31e-9
2.31e-9
</delay_matrix>
</pb_type>
<!-- Define the flip-flop -->
@ -462,20 +432,22 @@
<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_clock_to_Q max="124e-12" port="ff.Q" clock="clk"/>
<T_setup value="0.39e-9" port="ff.D" clock="clk"/>
<T_clock_to_Q max="0.43e-9" port="ff.Q" clock="clk"/>
</pb_type>
<interconnect>
<direct name="direct1" input="ble3.in[2:0]" output="lut3[0:0].in[2:0]"/>
<direct name="direct2" input="lut3[0:0].out" output="ff[0:0].D">
<!-- Advanced user option that tells CAD tool to find LUT+FF pairs in netlist -->
<pack_pattern name="ble3" in_port="lut3[0:0].out" out_port="ff[0:0].D"/>
<!-- Consider the delay of the MUX between LUT3 and FF -->
<delay_constant max="1.14e-9" in_port="lut3[0:0].out" out_port="ff[0:0].D"/>
</direct>
<direct name="direct3" input="ble3.clk" output="ff[0:0].clk"/>
<mux name="mux1" input="ff[0:0].Q lut3.out[0:0]" output="ble3.out[0:0]">
<!-- LUT to output is faster than FF to output on a Stratix IV -->
<delay_constant max="25e-12" in_port="lut3.out[0:0]" out_port="ble3.out[0:0]"/>
<delay_constant max="45e-12" in_port="ff[0:0].Q" out_port="ble3.out[0:0]"/>
<!-- Combine the delay of LUT4/LUT3 output MUX and fabric output mux -->
<delay_constant max="2.03e-9" in_port="lut3.out[0:0]" out_port="ble3.out[0:0]"/>
<delay_constant max="0.89e-9" in_port="ff[0:0].Q" out_port="ble3.out[0:0]"/>
</mux>
</interconnect>
</pb_type>
@ -505,20 +477,11 @@
<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
398e-12
397e-12
-->
<delay_matrix type="max" in_port="lut4.in" out_port="lut4.out">
261e-12
261e-12
261e-12
261e-12
2.6e-9
2.6e-9
2.6e-9
2.6e-9
</delay_matrix>
</pb_type>
<!-- Define flip-flop -->
@ -526,20 +489,22 @@
<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_clock_to_Q max="124e-12" port="ff.Q" clock="clk"/>
<T_setup value="0.39e-9" port="ff.D" clock="clk"/>
<T_clock_to_Q max="0.43e-9" port="ff.Q" clock="clk"/>
</pb_type>
<interconnect>
<direct name="direct1" input="ble4.in" output="lut4[0:0].in"/>
<direct name="direct2" input="lut4.out" output="ff.D">
<!-- Advanced user option that tells CAD tool to find LUT+FF pairs in netlist -->
<pack_pattern name="ble4" in_port="lut4.out" out_port="ff.D"/>
<!-- Consider the delay of the MUX between LUT4 and FF -->
<delay_constant max="1.14e-9" in_port="lut4.out" out_port="ff.D"/>
</direct>
<direct name="direct3" input="ble4.clk" output="ff.clk"/>
<mux name="mux1" input="ff.Q lut4.out" output="ble4.out">
<!-- LUT to output is faster than FF to output on a Stratix IV -->
<delay_constant max="25e-12" in_port="lut4.out" out_port="ble4.out"/>
<delay_constant max="45e-12" in_port="ff.Q" out_port="ble4.out"/>
<!-- Combine the delay of LUT4/LUT3 output MUX and fabric output mux -->
<delay_constant max="2.03e-9" in_port="lut4.out" out_port="ble4.out"/>
<delay_constant max="0.89e-9" in_port="ff.Q" out_port="ble4.out"/>
</mux>
</interconnect>
</pb_type>
@ -561,15 +526,27 @@
<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_clock_to_Q max="124e-12" port="ff.Q" clock="clk"/>
<T_setup value="0.39e-9" port="ff.D" clock="clk"/>
<T_clock_to_Q max="0.43e-9" port="ff.Q" clock="clk"/>
</pb_type>
<interconnect>
<direct name="direct1" input="shift_reg.reg_in" output="ff[0].D"/>
<direct name="direct2" input="ff[0].Q" output="ff[1].D"/>
<direct name="direct1" input="shift_reg.reg_in" output="ff[0].D">
<!-- Consider the delay of the MUX between LUT4 and FF -->
<delay_constant max="1.12e-9" in_port="shift_reg.reg_in" out_port="ff[0].D"/>
</direct>
<direct name="direct2" input="ff[0].Q" output="ff[1].D">
<!-- Consider the delay of the MUX between LUT4 and FF -->
<delay_constant max="0.56e-9" in_port="ff[0].Q" out_port="ff[1].D"/>
</direct>
<direct name="direct3" input="ff[1].Q" output="shift_reg.reg_out"/>
<direct name="direct4" input="ff[0].Q" output="shift_reg.ff_out[0:0]"/>
<direct name="direct5" input="ff[1].Q" output="shift_reg.ff_out[1:1]"/>
<direct name="direct4" input="ff[0].Q" output="shift_reg.ff_out[0:0]">
<!-- Consider the delay of the MUX between LUT4 and FF -->
<delay_constant max="0.88e-9" in_port="ff[0].Q" out_port="shift_reg.ff_out[0:0]"/>
</direct>
<direct name="direct5" input="ff[1].Q" output="shift_reg.ff_out[1:1]">
<!-- Consider the delay of the MUX between LUT4 and FF -->
<delay_constant max="0.89e-9" in_port="ff[1].Q" out_port="shift_reg.ff_out[1:1]"/>
</direct>
<complete name="complete1" input="shift_reg.clk" output="ff.clk"/>
</interconnect>
</pb_type>
@ -591,52 +568,36 @@
I[0] should be connected to in[0]
-->
<direct name="direct_fle0" input="clb.I0[0:2]" output="fle[0:0].in[0:2]">
<!-- TODO: Timing should be backannotated from post-PnR results -->
</direct>
<direct name="direct_fle0i" input="clb.I0i" output="fle[0:0].in[3]">
<!-- TODO: Timing should be backannotated from post-PnR results -->
</direct>
<direct name="direct_fle1" input="clb.I1[0:2]" output="fle[1:1].in[0:2]">
<!-- TODO: Timing should be backannotated from post-PnR results -->
</direct>
<direct name="direct_fle1i" input="clb.I1i" output="fle[1:1].in[3]">
<!-- TODO: Timing should be backannotated from post-PnR results -->
</direct>
<direct name="direct_fle2" input="clb.I2[0:2]" output="fle[2:2].in[0:2]">
<!-- TODO: Timing should be backannotated from post-PnR results -->
</direct>
<direct name="direct_fle2i" input="clb.I2i" output="fle[2:2].in[3]">
<!-- TODO: Timing should be backannotated from post-PnR results -->
</direct>
<direct name="direct_fle3" input="clb.I3[0:2]" output="fle[3:3].in[0:2]">
<!-- TODO: Timing should be backannotated from post-PnR results -->
</direct>
<direct name="direct_fle3i" input="clb.I3i" output="fle[3:3].in[3]">
<!-- TODO: Timing should be backannotated from post-PnR results -->
</direct>
<direct name="direct_fle4" input="clb.I4[0:2]" output="fle[4:4].in[0:2]">
<!-- TODO: Timing should be backannotated from post-PnR results -->
</direct>
<direct name="direct_fle4i" input="clb.I4i" output="fle[4:4].in[3]">
<!-- TODO: Timing should be backannotated from post-PnR results -->
</direct>
<direct name="direct_fle5" input="clb.I5[0:2]" output="fle[5:5].in[0:2]">
<!-- TODO: Timing should be backannotated from post-PnR results -->
</direct>
<direct name="direct_fle5i" input="clb.I5i" output="fle[5:5].in[3]">
<!-- TODO: Timing should be backannotated from post-PnR results -->
</direct>
<direct name="direct_fle6" input="clb.I6[0:2]" output="fle[6:6].in[0:2]">
<!-- TODO: Timing should be backannotated from post-PnR results -->
</direct>
<direct name="direct_fle6i" input="clb.I6i" output="fle[6:6].in[3]">
<!-- TODO: Timing should be backannotated from post-PnR results -->
</direct>
<direct name="direct_fle7" input="clb.I7[0:2]" output="fle[7:7].in[0:2]">
<!-- TODO: Timing should be backannotated from post-PnR results -->
</direct>
<direct name="direct_fle7i" input="clb.I7i" output="fle[7:7].in[3]">
<!-- TODO: Timing should be backannotated from post-PnR results -->
</direct>
<complete name="clks" input="clb.clk" output="fle[7:0].clk">
</complete>
@ -650,7 +611,7 @@
<!-- Shift register chain links -->
<direct name="shift_register_in" input="clb.reg_in" output="fle[0:0].reg_in">
<!-- Put all inter-block carry chain delay on this one edge -->
<delay_constant max="0.16e-9" in_port="clb.reg_in" out_port="fle[0:0].reg_in"/>
<delay_constant max="0e-9" in_port="clb.reg_in" out_port="fle[0:0].reg_in"/>
<!--pack_pattern name="chain" in_port="clb.reg_in" out_port="fle[0:0].reg_in"/-->
</direct>
<direct name="shift_register_out" input="fle[7:7].reg_out" output="clb.reg_out">
@ -662,7 +623,7 @@
<!-- Scan chain links -->
<direct name="scan_chain_in" input="clb.sc_in" output="fle[0:0].sc_in">
<!-- Put all inter-block carry chain delay on this one edge -->
<delay_constant max="0.16e-9" in_port="clb.sc_in" out_port="fle[0:0].sc_in"/>
<delay_constant max="0e-9" in_port="clb.sc_in" out_port="fle[0:0].sc_in"/>
</direct>
<direct name="scan_chain_out" input="fle[7:7].sc_out" output="clb.sc_out">
</direct>