Protecting SAPHO: synthesis without showing the source

Status: notes, 2026-10-14. Nothing decided, nothing built. Parked until a partnership between UFJF and a company asks for it (Luciano). Marks: fact (read or checked), from the docs (vendor documentation or forum answers found on 2026-10-14, quoted through a search where the page itself could not be opened), deduced.

1. The question

yanc writes a readable .v: a top module that instantiates processor with the program’s parameters (which operations to build, memory sizes, widths) and names its two .mif. To put it in an FPGA the synthesis tool also needs SAPHO/. The question: hand SAPHO/ to Quartus or Vivado so it synthesizes, without anyone being able to read it, the way vendors ship their own soft cores and IP. Simulation is a separate question, left out here (section 6).

2. What does not fit SAPHO

A pre-synthesized netlist (a Quartus exported partition, .qxp in Standard or .qdb in Pro; a Vivado checkpoint). Deduced: it freezes the parameters. SAPHO is generated per program (pay only for what you use), so one netlist serves one program. It fits only as a service (section 5).

3. Encrypted source, IEEE 1735

How it works (fact, the files in Quartus Lite 24.1 show the structure: key_method = "rsa", data_method = "aes128-cbc"): the IP is encrypted once with a random AES key; that key is wrapped once per tool with the tool vendor’s RSA public key; the tool holds the private key in its executable, decrypts in memory, applies the parameters and synthesizes; the decrypted source never reaches the disk. Deduced: the parameters keep working, so yanc’s .v would not change.

Quartus.

  • From the docs: the Quartus Prime Pro Edition guide says Pro supports IEEE 1735 v1 decryption and that Standard does not; Pro ships the public key (quartus/common/misc/public_key) and an encryptor encrypt_1735. Nothing direct was found for Lite, which sits below Standard.

  • Fact (this machine, Quartus Lite 24.1): no public_key, no encrypt_1735. Its 9 012 IEEE 1735 files carry keys for simulators only (Siemens MTI 5 595, Aldec 3 759, Synopsys VCS 2), none for Quartus: Intel’s own IP for synthesis in Lite uses another scheme.

  • From the docs (forum, Intel staff): for Standard there is Intel’s proprietary encryption, Ampcrypt/Amplic, obtained through Intel sales.

  • The catch for the lab’s boards: as far as we know, Cyclone V (the DE10-Nano) is supported by Lite and Standard only, not by Pro, so IEEE 1735 is not a path there. To be confirmed in Intel’s device support list.

Vivado.

  • From the docs (AMD UG835, UG1118): the Tcl command encrypt -lang verilog -key <keyfile> ... encrypts in IEEE 1735; it needs an encryption license, and the Xilinx public key is given to members of the IEEE P1735 group or through an AMD representative.

  • Not tested: no key and no license here.

Strength. From the docs: a 2021 paper (arXiv 2112.04838) reports key recovery from IEEE 1735 implementations of several vendors, Intel among them, and calls the weaknesses structural. Encryption keeps out casual readers and companies that will not risk a lawsuit; it is not a vault.

4. What was already published

Fact: SAPHO up to v7.1 is public, in the yanc repository and in every release, under the MIT license, and described in the IEEE paper. Those versions stay free for whoever has a copy. Only future versions could be protected, and only if their source leaves the public repository. Deduced: a patent on what the paper describes is unlikely; what UFJF could license is the future versions, the group’s know-how, the compiler and support. Dual licensing (a copyleft free version plus a paid commercial one) is possible only from a new version on.

5. Ways a partnership could deliver SAPHO closed

  1. Through the FPGA vendor: IEEE 1735 (Quartus Pro, Vivado with the AMD key and license) or Intel’s Ampcrypt for Standard/Lite. Needs an agreement with each vendor; yanc unchanged.

  2. As a service: the client sends the program (or yanc’s .v); the holder of SAPHO synthesizes that processor and returns a netlist for the client’s family (a Quartus partition, a Vivado checkpoint). Solves the per-program parameters; needs the holder to run the synthesis for each program and each FPGA family.

  3. Obfuscation only: renamed signals, no comments, one flat file. Works in every tool, simulators included; delays a reader, does not stop one.

6. Left out here

Simulation of a closed SAPHO: Icarus and Verilator cannot decrypt IEEE 1735 (open tools cannot hold the vendors’ private keys); commercial simulators can, but the inside of the processor is hidden in the waveform. An instruction-set simulator in yanc (issue #32) would let users run and debug programs without the HDL.

7. To do if a partnership comes

  • Ask the vendors: Intel (Ampcrypt/Amplic for Standard/Lite and the Cyclone V question; IEEE 1735 key for Pro), AMD (the IEEE 1735 key and encryption license).

  • With a key in hand: encrypt a small module, then SAPHO, and synthesize the yanc output of sapho_all; compare Fmax and area with the plain source.

  • Decide the licence of future versions with UFJF’s technology transfer office before any SAPHO source leaves the public repository.

Sources (2026-10-14): Quartus Prime Pro Edition User Guide, “Support for the IEEE 1735 Encryption Standard” (683463, 24.1 and 25.1); Altera Community threads “Quartus Prime Standard encryption public key”, “Decrypt problem in Std version of Quartus”, “How to use IEEE 1735 encryption on a design with Cyclone V?”; Macnica FAQ 141642; AMD UG835 encrypt, UG1118 “Encrypting IP with Vivado”; arXiv 2112.04838.