Repository navigation
Silinx 15.24.0
What's new in 15.24.0
The 7-series arrive, as an experimental family: the Basys 3 and Nexys A7 boards, implemented by AMD Vivado as
the Spartan-3E is by ISE. The Spartan-6's nextpnr places the block RAMs and DSPs itself, with shorter periods, and
several of its Map and Place & Route failures are fixed.
In short:
- The nextpnr engine's placements a little faster: the placer knows that a LUT on a SLICEM is slower than one on a
SLICEL, and keeps the critical paths' LUTs off the SLICEMs. - Spartan-6 (Nexys3, Atlys): a design whose outputs have no LOC routes again. Place & Route had failed on every seed
when Silinx put such an output on a pad of the top bank that nextpnr cannot reach. - Boards: the Digilent Basys 3 and Nexys A7 (the 7-series' Artix-7), an experimental family: their pins, their
constraints written as an XDC; no flow implements them yet. - Spartan-6 (Nexys3, Atlys): Place & Route gives shorter clock periods, about 3 % on average over the examples and
up to 15 %. nextpnr now places the block RAMs and DSP48A1s itself. - The 3D Karnaugh map of 3 and 4 inputs: every cell in view shows its number and value on the cylinder and the
torus, smaller where the surface is seen at a slant, never beside it. A group's outline on cells hidden behind the
surface is no longer drawn over the surface in front of them. - A short click on an emulated push button counts: the Board Emulator holds a button down until the design has
seen a real press, so the clock, pong, stopwatch and blinky examples' debounced buttons react to a short click,
not only to a long one. - Help ▸ About Silinx names the open toolchain's boards of today (Basys2, Nexys2, Nexys3, Atlys, iCEBreaker)
and its engines (Yosys, nextpnr, the bitstream writer of Silinx or icepack). - Smaller designs that compare a signal with several constants: such comparisons are made of LUTs instead of carry
chains (the vga example 143 → 122 slices, 10.2 → 9.2 ns). - AMD Vivado for the Artix-7 boards: a project on the Basys 3 or the Nexys A7 is synthesized, implemented and
made into a bitstream by AMD Vivado ML Standard 2025.2, in its Docker image, over ssh or on this computer, as Xilinx
ISE is for the Spartan-3E: its processes in the Processes pane, its utilization, timing and design rule check in the
Design Summary, its warnings in the console, the board programmed with openFPGALoader. - Synthesis: in a design of several clocks, each clock's critical path is made shorter, not only the slowest clock's.
In detail:
- test/compiled-words.test.js: the word-level models of the stopwatch, the clock and the vga examples expected as #3097's
comparisons as logic (384b745) synthesize them (#3161): their cells and their tables' counts; the test failed on main
since. - The Spartan-3E's nextpnr engine at 1.4.2 (#3157). nextpnr-himbaechel-spartan3e 1.4.2 (nextpnr-spartan PR 30,
40a7f9b3): the placer's estimate of a connection from a SLICEM's LUT is 55 ps longer (Tilo 0.759 ns against a SLICEL's
0.704 at -4,fpga/timing.jsSPEED_4); the chip database's LUT4 has a SLICEL's arc on every site. Natively, XC3S1200E,
the mean period of one placement per seed: fpu-adder 10.975 -> 10.714 ns (-2.4 %, seeds 1-4), iwls05-sasc 4.780 ->
4.731 ns (-1.0 %) and iwls05-simple-spi 6.548 -> 6.478 ns (-1.1 %, seeds 1-16). Part of #2982, which stays open. - The nextpnr and Yosys update comparisons measure the base's package on the base's side (#3156):
nextpnr-update.ymlandyosys-update.ymlinstalled the base's package at the repository's root only, and npm kept
the pull request's nested inpackages/silinx-fpga/node_modules, which the toolchain resolves first: both sides
measured the pull request's package. The base's is now installed into the toolchain's workspace and the root both,
and a step on each side prints the version resolved frompackages/silinx-fpgaand from the root and fails when it
is not the side's (scripts/resolved-package.mjs). - A port without a LOC takes a pad whose I/O path is known (#3152). nextpnr's chip database leaves out the PIPs
whose bits Silinx's bit database does not know. One of them is the slave OLOGIC2's D1 -> OQ route-through in the top
bank's TIOI_INNER tiles (the master's is known), so 30 pads of the XC6SLX16 and 23 of the XC6SLX45 can take no
output, and 4 pads of each can take no input. Silinx's placer gave such pads to ports without a LOC, and nextpnr
failed on every seed:Failed to route arc 0.0 of net …, from X6Y4/L_AQ to X5Y0/TIOB_O3_PINW. The new module
fpga/s6/pad-paths.js(padPathsS6) finds each pad's paths through the known PIPs, in about 0.1 s. Probed
pad by pad, nextpnr agrees on both parts. The placer now gives ports without a LOC only pads with their path. The examples, all
of their ports LOC'd, are placed as before, so their bitstreams are unchanged. - The Spartan-6's Map packs a product of two counters (#3151): it stopped with "its pin BMUX carries two nets"
when a carry chain's sums left their slice (to the DSP48A1s) by the multiplexed outputs a 5FF had been given too;
such a flip-flop now goes to another position (the toolchain's fragment has the detail). - The Basys 3 and the Nexys A7 boards, their constraints as XDC (#3142): the Digilent Basys 3 (xc7a35tcpg236-1) and
Nexys A7 (xc7a100tcsg324-1) in the board choice of New Project and Design Properties, marked "(experimental)", as is
their family, Artix-7: their clock, LEDs, switches, buttons, seven-segment display, VGA, USB-UART, PS/2 and Pmods,
from Digilent's master XDCs (every pin checked against them). A project of the family has its constraints in
constraints/top.xdc: Update Constraints, New Source and an example made for the board write the board's pins as
Vivado reads them (each pin'sPACKAGE_PINandIOSTANDARD, the clock'screate_clock, the board'sCFGBVSand
CONFIG_VOLTAGE), and Check Constraints reads it. No flow runs the family yet (Vivado #1649, the Silinx toolchain
#3146): its process tree has the constraints and the checks only, the synthesis tool is locked, Implement and
Program are not offered, and the console says so when the board is chosen. The XC7A35T joins the device database;
Artix-7's constraints were UCFs for ISE before. The MCP server'slist_devicesmarks the family and its boards
experimental,set_design_properties,new_project,new_sourcewrite XDC andrun_processrefuses with the
interface's message.new_sourcenow writes an iCE40's constraints with the PCF template (it wrote the UCF's). - The Spartan-6's nextpnr engine: nextpnr-himbaechel-spartan6 0.2.0, which places the block RAMs and DSP48A1s (#3137).
Its chip database has each routing node's delay from Silinx's timing model (#3149). The annealing refinement after
its analytical placer reaches 40 tiles, so a block RAM or DSP48A1 moves next to its logic. Silinx no longer places
them and fixes them for nextpnr (S6_HARD_BY_NEXTPNRtrue). The examples' periods (Optimization Speed, best of
3 seeds) are 3.2 % shorter on average on the Nexys3 and the Atlys, with every PERIOD met; for example the stopwatch on
the Nexys3 went from 9.959 to 8.941 ns and micro80 on the Atlys from 8.843 to 8.613 ns. - The Spartan-6's nextpnr engine: nextpnr-himbaechel-spartan6 0.3.0 (#3135, #3136). The package can pack Yosys's
netlist into sites and place its pads and global buffers as Silinx's Map and placer do
(pmnmalo/nextpnr-spartan#29); Silinx's flow still hands it its own Map's sites, placed and routed as before (#3138
will hand it Yosys's netlist). - The unit tests' timings refreshed (#3131): test/timings.json from v15.23.0's release run, the first with five Linux
parts (#3124): the split files' measured times replace their estimated shares, the parts balanced again. - The private-data check allows #2962's special-purpose addresses in #3123's commit message too (#3129): that
message names two of the reserved ranges, and the history check of the release's CI failed on it. - A release's CI in about 45 minutes instead of 80 to 120 (#3124): the unit tests in five parts on Linux, all of
them; on macOS and Windows in three parts, but the toolchain's heaviest (HEAVYin scripts/test-shard.mjs, its new
option--light: the whole open flow on every feature, device and example, the post-place & route and Spartan-6
models of the examples, the synthesis on the corpus, nextpnr's engine), about half the time, which took their parts
to 60 to 80 minutes there; test/open-flow.test.js (blinky to a bitstream, the shipped devices) stays their smoke test.
The runs of the tags one at a time: the IDE's and the toolchain package's tags pushed together shared the 20 jobs a
public repository runs at once and queued them up to 50 minutes (v15.22.0). test/timings.json refreshed from
15.23.0's run (schbeautify-sheet.test.js weighed 30 seconds, took 22 minutes). CLAUDE.md says so. - The two longest unit test files split (#3124): test/open-flow.test.js (34 to 50 minutes in CI, its tests one after
another) into open-flow.test.js (the shipped device descriptions, blinky), open-flow-features.test.js (block RAMs,
multipliers, shift registers), open-flow-io.test.js (DCMs, global buffers, IOBs) and open-flow-clocks.test.js, their
helpers and fixtures' table in test/open-flow-lib.mjs; test/sim-model-par.test.js (21 to 38 minutes) into
sim-model-par.test.js, sim-model-par-examples.test.js and sim-model-par-micro80.test.js. The same tests, moved as they
were; test/timings.json gives the new files their share of the old ones' times. - The private-data check allows #2962's special-purpose addresses in CHANGELOG.md (#3121): the NAT64 and reserved
ranges that a link's zip address is checked against were allowed in their fragment only, so the release that moved
the fragment into CHANGELOG.md failed its CI. - The cells' texts on the 3D map of 3 and 4 inputs (#3119): a cell's text is shown wherever the cell faces the
viewer. That is its centre while the centre faces the viewer well, else the part of the cell that does (before, the
centre's normal had to point at least a quarter towards the screen, which hid cells in plain view: on the torus only
7, 11 and 14 showed a value). The text is upright and shrinks with the surface's foreshortening. It is drawn after
its own cell's faces (the cell's nearer faces covered it), only where its box lies on the surface and nothing nearer
hides it, shrunk or moved within the cell when it does not fit. The 0s of an SOP map stay blank, as on the flat map
(core/kmapsurface.js: labelSpots, textFits, textMatrix, surfaceFrame). - The group outlines of the 3D map of 3 and 4 inputs (#3120): what lies on the faces seen from behind is drawn
first: on the closed torus such a face is always behind a face seen from the front, and on the open cylinder it is
the inside. On the torus it is not drawn at all. A sort by depth alone drew, along the silhouette, the outlines of
hidden groups over the surface in front (the torus's m(4, 5), the cylinder's a.c). An outline segment is drawn after
the faces round it, so a neighbouring face no longer cuts it into dashes (core/kmapsurface.js: paintOrder,
kmapPatch'saround). Tests: kmapsurface-view (the groups' cells against the flat map's, the texts' spots, size and
fit, the order on the torus in six views); ui/tt-kmapsurface-view (the two maps of the issues, in two views). - docs/OPEN-TOOLCHAIN.md without its "Open questions for the legal review" (#3114): the section is removed; its
facts on the device descriptions (shipped for the supported boards' six devices, the Spartan-3E dies also measured
black-box) are a paragraph of the method, no longer saying the Spartan-6 toolchain is under development. The
introduction covers the Spartan-6 too and no longer asks for a review before publishing; the research notes of
s3e-route and s3e-device no longer cite an open legal question, and s3e-route's says the computed descriptions ship. - Push buttons held as long as a real press (#3109): a design's debouncer takes a press once the button has
stayed down for some milliseconds of the board's time (the examples': 20 ms, a million cycles), and the emulator,
slower than real time (the event-driven engine much slower), let a short click go after fewer cycles. A button let
go now stays down, lit, until 100 ms of the board's time have passed since it was pressed (at a set clock rate,
100 ms of its cycles, at least one); a longer press is let go at once, a button let go while paused at once, and
Shift+click still keeps it pressed (core/emulate-press.js). - The About's open toolchain (#3108): the About dialog said the Silinx toolchain covered the Basys2 and Nexys2
only, with a place and route of its own for the Spartan-3E; it now names the five boards, the Lattice iCE40 beside
the Xilinx FPGAs, and nextpnr, in English and in Portuguese. - A signal compared with several constants as LUTs (#3097): the Silinx toolchain's synthesis maps the comparisons with
a constant of a signal compared with two constants or more, and those that test the signal's high bits (count > 0),
as logic before Yosys's coarse step, not as carry chains whose LUTs Place & Route had to copy for each chain (itc99-b05
252 → 194 slices, verilog-uart 117 → 111 slices and 8.7 → 6.9 ns); silinx-fpga's changelog has the details. - The old name split by a line break (#2624, part of the rename to Silinx):
test/old-name.test.jscounts the
old name when a line break comes between its two words (in a comment, a Markdown quote or list, or plain text),
which it missed; the one such place left, a comment of the silinx-fpga package's command line, says Silinx. - The ISE runner's names say Silinx (#2624): its launchd daemons and agents are
com.github.pmnmalo.silinx.*,
its scripts' folderLibrary/Silinx-runner;scripts/ise-runner/install.shunloads and removes the old ones. The
ISE tests' work folders on the ISE host aresilinx-examples,silinx-testsandsilinx-timesim. The old-name
guard lists only what stays on purpose: the old commandbin/silinx-ise.js(removed later, #3133) and the runner's
clean-up of its old names. - Vivado as the Artix-7's back end (#1649):
server/vivado.jswrites a batch Tcl flow in non-project mode
(synth_design,opt_design,place_design,route_design,write_bitstream, each step's checkpoint, the
reports), with ISE's step names so the Processes and the MCP server's processes run it unchanged; CFGBVS and
CONFIG_VOLTAGE set unless the board's XDC sets them (no DRC CFGBVS-1). Toolchain Settings'vivado(its mode:
docker, ssh or local; its image; its settings64.sh), its container with ISE's isolation and sysfs's buses, classes
and devices hidden (Vivado's licence manager crashes without, #2305).server/vivado-reports.jsparses
report_utilization, report_route_status, report_timing_summary, report_drc, report_power and the log (each
command's own messages) into the Design Summary's shape. The interface: the Artix-7's tool descriptor (VIVADO:
Synthesize - Vivado, the implementation, the timing, the power, the pins, the programming file; no simulation
model, no Emulate), a run without Vivado configured says so and writes the scripts; the browser edition keeps
#3142's "no implementation flow yet". The MCP server's run_process (toolchain vivado) and get_report (Vivado's
reports, a drc report). The other 7-series families keep ISE's flow. - The examples on the Basys 3 through Vivado (#1649):
test/vivado/examples.test.js(SILINX_VIVADO_TESTS=1, on
the ISE machine): all ten routed, the 10 ns clock met, a .bit for the xc7a35t, the design rule check with no error;
the warnings asserted: none, but synthesis's on traffic's and pong's outputs driven by a constant and on micro80's
light8080 core's unread bits (#2544), and the design rule check's DSP pipelining advice on stopwatch. - MCP get_report's schema (#3158): its
reportargument lists the reports (summary, power, timing, synthesis,
map, par, pins, drc), not the toolchains. - Each clock's own critical path made shorter by the open synthesis (#652): a design of several clocks has the
critical LUTs of every clock made shallower, not only the slowest clock's. vtr-stereovision3'stm3_clk_v010.9 %
shorter at Speed by ISE's trce (5.958 -> 5.306 ns) and 9.0 % at Area, its counter's enable on 2 levels of LUTs as
XST's, with the same size; the toolchain's package's changelog has the details.
Silinx is an ISE-style FPGA design environment that runs in the browser, made for teaching digital systems: VHDL/Verilog projects; schematics with live simulation; FSM and ASM state-machine editors, truth tables and Karnaugh maps, all kept in sync with their HDL; module and test bench wizards; behavioural and netlist simulation with waveforms; a board emulator; synthesis, place & route and bitstreams with its own toolchain (the Basys2 and Nexys2 boards) or with Xilinx ISE 14.7; and board programming, also from the browser.
Downloads
| File | What it is |
|---|---|
| Silinx.html | Single-file edition, the same app as the site (https://pmnmalo.github.io/Silinx/). Download it and open it in a browser: design, simulation, synthesis, place & route and bitstream (the Silinx toolchain, Basys2 / Nexys2; its tools come from the site the first time). Projects are kept in the browser (File ▸ Download Project Bundle to back them up) or, in Chrome / Edge, in a folder on the disk. Programming the board's FPGA and Platform Flash over USB works from the site's address (Chrome / Edge, Chrome on Android). |
| silinx-<version>.zip | Silinx run locally, with its dependencies: an alternative to the site (offline synthesis, projects as folders, the installed programmers, Xilinx ISE 14.7 modes). Needs Node.js 18 or newer. Unzip, then double-click Start Silinx.bat (Windows), Start Silinx.command (macOS) or start-silinx.sh (Linux) in the silinx folder; Silinx opens at http://127.0.0.1:8642. Projects are stored in ~/Silinx-projects. |
| silinx-docker-kit-<version>.zip | Kit to build your own private Docker image of Xilinx ISE 14.7 (Windows, macOS Intel/Apple Silicon, Linux) for synthesis, place & route and bitstreams. ISE and its licence are not included and must not be redistributed: download them with your own AMD account. See ise/README.md. |
The source code is attached by GitHub below (Source code zip / tar.gz).
Silinx is free software (GNU AGPL-3.0, Copyright 2026 Pedro Maló; see LICENSE, NOTICE and THIRD-PARTY-NOTICES.md). It is an independent project, not affiliated with or endorsed by AMD/Xilinx. Xilinx and ISE are trademarks of AMD.
Checking a download
The files' SHA-256 checksums (the file SHA256SUMS):
fee82d345c95bb1d23da6dd21c619399e184b97fb5c5e0fb0bcf149360926b9b Silinx.html
b0c2803fb9a2126625bda6441acaa23fe6c94702546add0640de9a86cb33d403 silinx-15.24.0.zip
f7486ca426e89de9de2146eb205f6dbb899cfb2dd412fd698b7ca16150ca95de silinx-docker-kit-15.24.0.zip
In the folder of the downloads: sha256sum -c SHA256SUMS --ignore-missing (Linux), shasum -a 256 -c SHA256SUMS --ignore-missing (macOS), Get-FileHash <file> (Windows PowerShell, compared by eye). Each file also has a build provenance attestation (it was built by this repository's Release workflow from this tag): gh attestation verify <file> --repo pmnmalo/Silinx.