107 ROTFPGA v2a

107 : ROTFPGA v2a

Design render
  • Author: htfab
  • Description: A reconfigurable logic circuit made of identical rotatable tiles
  • GitHub repository
  • Clock: 10000000 Hz

How it works

Logic tile

A reconfigurable logic circuit built from identical copies of the tile above containing a NAND gate, a D flip-flop and a buffer, with each tile individually rotated or reflected as described by the FPGA configuration. Port of the original ROTFPGA from Caravel to TinyTapeout.

Porting the design required a 50-fold decrease in chip area which was achieved using a combination of cutting corners, heavy optimization and a few design changes. In particular:

  • The FPGA was reduced from 24×24 to 8×8 tiles. There are 8 inputs and 8 outputs instead of 12 each.
  • To compensate for smaller size, tiles can also be mirrored in addition to rotation.
  • Tiles (being the most repeated part of the design) were rewritten as hand-optimized gate-level Verilog.
  • Each tile only contains 1 flip-flop (the one exposed to the user). Configuration is now stored in latches.
  • Configuration and reset are performed using a routing-efficient scan chain, so the design is no longer routing constrained. This allows standard cells to be placed with >80% density.
  • Openlane and its components are 2 years more mature, hardening the same HDL more efficiently.

Configuration

Each tile can be configured in 8 possible orientations. Bits 0, 1 and 2 correspond to a diagonal, horizontal and vertical flip respectively. Any rotation or reflection can be described as a combination:

Rotations and reflections (The bottom row looks somewhat different, but we just rearranged the wires so that the inputs and outputs line up with the unmirrored tiles.)

Tiles are arranged in an 8×8 grid:

  • Top, bottom, left and right inputs and outputs are connected to the tile in the respective direction.
  • Tiles mostly wrap around, e.g. the bottom output of a cell in the last line connects to the top input of the cell in the first line.
  • As an exception to the wrapping rules, left inputs in the first column correspond to chip inputs and right outputs in the last column correspond to chip outputs.
  • There is a scan chain meandering through all the tiles, visiting lines from top to bottom and within each line going from left to right.

This is a 4×4 model of the tile grid, showing regular i/o as black and the scan chain as grey:

Grid model

When the scan enable input is 0, the FPGA operates normally and each tile sets its flip-flop to the input it receives from one of the neighboring tiles according to its current rotation/reflection. When scan enable is 1, it sets the flip-flop to the value received through the scan chain instead. This allows us to set the initial state of each flip-flop and also to query their state later for debugging. With some extra machinery it also allows us to change the rotations/reflections.

When the 2-bit configuration input is is 01, each cell updates its vertical flip bit to the current value of its flip-flop. Similarly, for 10 it sets the horizontal flip and for 11 it sets the diagonal flip. When configuration is 00, all three flip bits are latched and the orientation doesn't change.

One can thus configure the FPGA by sending the sequence of all diagonal flip bits through the scan chain, then setting configuration to 11 and back to 00, then sending all horizontal flip bits, setting configuration to 10 and back to 00, and finally sending the vertical flip bits and setting configuration to 01 and back to 00.

Note that in order to save space the flip bits are stored in latches, not registers. Changing the configuration input from 00 to 11 or vice versa can cause a race condition where it is temporarily 01 or 10, overwriting the horizontal or vertical flip bits. Therefore one should configure the diagonal flips first.

Loop breaker

The user design may intentionally or inadvertantly contain combinational loops such as ring oscillators. To help debug such designs, the chip has a loop breaker mechanism using a loop breaker enable input as well as a 2-bit loop breaker class input.

Tiles are assigned to loop breaker classes:

Loop breaker tile classes

The loop breaker latches a tile output if and only if the following conditions are all met:

  • The loop breaker enable input is 1.
  • The current tile has a non-empty class that is different from the loop breaker class input.
  • The output doesn't come from the tile's flip-flop.

The loop breaker has the following properties:

  • If loop breaker enable is 1 and loop breaker class is constant, there are no combinational loops running. If we also pause the clock, the circuit keeps a steady state.
  • If loop breaker enable is 1 and we cycle loop breaker class through all possible values repeatedly while the clock is paused, everything will eventually propagate. If we also assume that the design has no race conditions, it will behave in the same way as if loop breaker enable was 0.

Reset

Setting the active-low reset input to 0 has the following effect:

  • Override scan enable to 1, scan chain input to 0 and disengage the latches for vertical, horizontal and diagonal flips. When kept low for 64 clock cycles this will reset the state and configuration in every tile.
  • Override loop breaker enable to 1 and loop breaker class to 00. This ensures that we play nice with other designs on TinyTapeout and keep a steady state while our design is not selected.

Pin mapping

Input pins:

  • clk provides a clock signal for the flip-flops
  • rst_n is the active-low reset described above
  • ui_in[7:0] are passed to the leftmost column of tiles as inputs from the left

Output pins:

  • uo_out[7:0] come from the rightwards output of the rightmost column of tiles

Bidirectional pins:

  • uio_in[0] is the scan enable input
  • uio_in[1] is the scan chain input
  • uio_in[3:2] are the configuration input bits
  • uio_in[4] is the loop breaker enable input
  • uio_in[6:5] are the loop breaker class input bits
  • uio_out[7] is the scan chain output

How to test

Follow the test suite the test directory. It sets up the FPGA with the following two configurations and runs a battery of tests on each.

Test configuration 1 used for upload, download, single-step and propagation tests:

Diagram corresponding to `fpga_config` in `test.py`

Test configuration 2 used for testing the loop breaker with manual and automatic cycles:

Diagram corresponding to `cfg` from the loop breaker test in `test.py`

External hardware

None

IO

#InputOutputBidirectional
0tile(0,0) left intile(7,0) right out_scan enable_ input
1tile(0,1) left intile(7,1) right out_scan chain_ input
2tile(0,2) left intile(7,2) right out_configuration_ input bit 0
3tile(0,3) left intile(7,3) right out_configuration_ input bit 1
4tile(0,4) left intile(7,4) right out_loop breaker enable_ input
5tile(0,5) left intile(7,5) right out_loop breaker class_ input bit 0
6tile(0,6) left intile(7,6) right out_loop breaker class_ input bit 1
7tile(0,7) left intile(7,7) right out_scan chain_ output

Chip location

Controller Mux Mux Mux Mux Mux Mux Mux Mux Mux Mux Mux Mux Mux Mux Mux Mux Mux Mux Mux Mux tt_um_chip_rom (Chip ROM) tt_um_factory_test (TinyTapeout Factory Test) tt_um_MichaelBell_tinyQV (TinyQV Risc-V SoC) tt_um_urish_silife_max (Game of Life 8x32 (siLife)) tt_um_vc32_cpu (VC 16-bit CPU) tt_um_tinytapeout_logo_screensaver (VGA Screensaver with Tiny Tapeout Logo) tt_um_htfab_rotfpga2 (ROTFPGA v2a) tt_um_htfab_latch_test (Latch test) tt_um_no_time_for_squares_tommythorn (No Time For Squares, IHP edition) tt_um_tommythorn_maxbw (Asynchronous Multiplier) tt_um_urish_simon (Simon Says memory game) tt_um_htfab_rotfpga2_ff (ROTFPGA v2b) tt_um_meiniKi_ttihp_fazyrv_exotiny (FazyRV-ExoTiny) tt_um_rebeccargb_hardware_utf8 (Hardware UTF Encoder/Decoder) tt_um_rebeccargb_universal_decoder (Universal Binary to Segment Decoder) tt_um_algofoogle_raybox_zero (raybox-zero TTIHP0p2 edition) tt_um_z2a_rgb_mixer (RGB Mixer demo) tt_um_vga_clock (VGA clock) tt_um_frequency_counter (Frequency counter) tt_um_brandonramos_VGA_Pong_with_NES_Controllers (VGA Pong with NES Controllers) tt_um_demosiine_sda (DemoSiine) tt_um_toivoh_demo_deluxe (Sequential Shadows Deluxe [TT08 demo competition]) tt_um_edwintorok (Rounding error) tt_um_2048_vga_game (2048 sliding tile puzzle game (VGA)) tt_um_hpretl_spi (SPI Test) tt_um_top_mole99 (One Sprite Pony) tt_um_urish_spell (SPELL) tt_um_rebeccargb_vga_pride (VGA Pride) tt_um_autosel (I2C EEPROM Project Selection) tt_um_a1k0n_nyancat (VGA Nyan Cat) tt_um_a1k0n_vgadonut (VGA donut) tt_um_rebeccargb_colorbars (Color Bars) tt_um_crispy_vga (Crispy VGA) tt_um_kbeckmann_flame (Flame demo) tt_um_jamesrosssharp_1bitam (1bit_am_sdr) tt_um_simon_cipher (simon_cipher) tt_um_htfab_bouncy_capsule (Bouncy Capsule) tt_um_phansel_laplace_lut (Experiment Number Six: Laplace LUT) tt_um_kianv_bare_metal (KianV RISC-V RV32E Baremetal SoC) tt_um_calonso88_rsa (8 bit RSA encryption) tt_um_silice (Warp) tt_um_rejunity_vga_test01 (VGA Drop (audio/visual demo)) tt_um_a1k0n_demo (Demo by a1k0n) tt_um_MichaelBell_canon (TT08 Pachelbel's Canon demo) tt_um_htfab_caterpillar (Simon's Caterpillar) tt_um_ravenslofty_chess (Chess) tt_um_fountaincoder_top_V2 (maddihp) tt_um_tomkeddie_a (VGA Experiments in Tennis) tt_um_MichaelBell_mandelbrot (VGA Mandelbrot) tt_um_MichaelBell_rle_vga (RLE Video Player) tt_um_jayjaywong12 (mulmul) tt_um_wokwi_392873974467527681 (PILIPINASLASALLE) tt_um_froith_goldcrest (Goldcrest RISC-V) tt_um_dvxf_dj8v (DJ8 8-bit CPU) tt_um_hpretl_minilogix (Minilogix) tt_um_tomkeddie_b (Transmit UART) tt_um_joerdsonsilva_modem (Multimode Modem) tt_um_oled_frequency_counter (Frequency Counter SSD1306 OLED) tt_um_stochastic_addmultiply_CL123abc (Stochastic Multiplier, Adder and Self-Multiplier) tt_um_QIF_8bit (8 Bit Digital QIF) tt_um_toivoh_retro_console (Retro Console) tt_um_cejmu (CEJMU Beers and Adders) tt_um_rejunity_sn76489 (Classic 8-bit era Programmable Sound Generator SN76489) tt_um_dlmiles_tt05_i2c_bert (I2C BERT) tt_um_dlmiles_muldiv8 (MULDIV unit (8-bit signed/unsigned)) tt_um_dlmiles_loopback (IHP loopback tile with input skew measurement) tt_um_dlmiles_bad_synchronizer (Example of Bad Synchronizer) tt_um_wokwi_407306064811090945 (DDR throughput and flop aperature test) tt_um_urish_giant_ringosc (Giant Ring Oscillator (3853 inverters)) tt_um_digital_clock_example (Digital Desk Clock v2.0) tt_um_rejunity_z80 (Zilog Z80) tt_um_rejunity_ay8913 (Classic 8-bit era Programmable Sound Generator AY-3-8913) tt_um_rtfb_collatz (Collatz conjecture brute-forcer) tt_um_ccattuto_conway (Conway's Game of Life on UART and VGA) tt_um_snow (Snow) tt_um_calonso88_74181 (8-bit ALU based on 2x 74181) tt_um_rejunity_vga_logo (VGA Tiny Logo (1 tile)) tt_um_NicklausThompson_SkyKing (SkyKing Demo) tt_um_htfab_cells (Cell mux) tt_um_htfab_pg_1x1 (Power gating test (1x1)) tt_um_htfab_pg_1x2 (Power gating test (1x2)) tt_um_dlmiles_ringosc_5inv (Ring Oscillator (5 inverter)) tt_um_devinatkin_pulse_width_counter (Pulse Width Counter) tt_um_algofoogle_vga_fun_wrapper (TTIHP VGA FUN!) tt_um_cfib_demo (cfib Demoscene Entry) tt_um_vga_glyph_mode (Glyph Mode) tt_um_favoritohjs_scroller (VGA Scroller) tt_um_pulse_generator (TTL Pulse Generator) tt_um_rajum_iterativeMAC (Iterative MAC) tt_um_algofoogle_tinyvga_fun_wrapper (TTIHP TinyVGA FUN!) tt_um_urish_sram_test (SRAM (1024x8) test) tt_um_one_bit_puf_wrapper (One Bit PUF) tt_um_multi_bit_puf_wrapper (One Bit PUF) tt_um_gray_sobel (Gray scale and Sobel filter) tt_um_rebeccargb_intercal_alu (INTERCAL ALU)