Tiny Tapeout > Tiny Tapeout Chips > Tiny Tapeout IHP 0p4 > 106 Spongent-88 Hash Accelerator

106 Spongent-88 Hash Accelerator

106 : Spongent-88 Hash Accelerator

Author: Stefan Aeschbacher
Description: Hardware accelerator for the Spongent-88/80/8 lightweight hash function.
GitHub repository
Open in 3D viewer
Clock: 50000000 Hz

How it works

This chip implements the Spongent-88/80/8 lightweight hash function as a hardware accelerator, designed as the cryptographic primitive for Winternitz One-Time Signatures (W-OTS) — a post-quantum secure signature scheme.

Spongent-88/80/8

Spongent is a sponge-based hash function optimised for extremely constrained hardware (Bogdanov et al., CHES 2011). The 88/80/8 variant has:

88-bit internal state, split into an 8-bit rate and 80-bit capacity
45 permutation rounds per absorption step
A round function of: round counter injection → S-box layer → bit permutation (pLayer)

Each round applies 22 parallel 4-bit S-box lookups, a zero-gate bit permutation P(i) = (i × 22) mod 87, and XORs a 6-bit LFSR counter into both ends of the state (forward into bits [5:0], bit-reversed into bits [87:82]).

The permutation is implemented with 2-round unrolling: two full rounds per clock cycle (22 double-round cycles + 1 single-round cycle for round 44), giving a latency of exactly 23 cycles per permutation call.

Sponge construction

The host absorbs message bytes one at a time: each byte is XORed into the rate portion (state[7:0]) and the permutation is triggered. After all message bytes plus padding are absorbed, the full 88-bit state is the digest. Padding follows the pad10*1 rule (single byte 0x01 to set the first pad bit, 0x80 to set the last — for a byte-aligned message this collapses to 0x81).

W-OTS use case

W-OTS with Winternitz parameter w=16 uses 25 hash chains of depth up to 15. Each chain step is one Spongent-88 call. The chip accelerates all 25 × 15 = 375 permutations needed to sign a message; at 50 MHz this takes approximately 190 µs per signature (25 cycles × 375 calls at 50 MHz). Key management and protocol logic run in software on the host (RP2040).

Register interface

The chip is controlled through a simple byte-serial register interface over the TinyTapeout bidirectional pins:

Signal	Direction	Description
`ui_in[7:0]`	input	data byte to write
`uo_out[7:0]`	output	current digest byte (LSB-first)
`uio[2:0]`	input	register address
`uio[3]`	input	write strobe (rising-edge triggered)
`uio[4]`	input	read strobe — advances output byte at addr 2
`uio[0]`	output	busy — high while permutation is running
`uio[1]`	output	out_valid — high after squeeze until next reset

Register map:

Addr	Direction	Action
0	write `0`	Reset: zero the sponge state, clear `out_valid`
0	write `1`	Squeeze: latch 88-bit digest into output shift register
0	write `2`	Hash: absorb pad byte `0x81` then auto-squeeze (no manual padding needed)
1	write `b`	Absorb: XOR byte `b` into `state[7:0]`, run 45-round permutation
2	read strobe	Advance output shift register to next digest byte

Timing: one absorb call takes 25 clock cycles (1 load + 23 permutation rounds + 1 capture). The host must poll busy before issuing the next command.

How to test

Using the RP2040 demo board

Connect the TinyTapeout demo board. The chip runs at 50 MHz.

Hashing a message:

import machine, time

# Pin assignments (TinyTapeout demo board)
# ui_in  → 8 GPIO pins driving the data byte
# uio_in → 5 GPIO pins: [4]=rd_en, [3]=wr_en, [2:0]=addr
# uio_out → 2 GPIO pins: [1]=out_valid, [0]=busy
# uo_out → 8 GPIO pins for reading digest bytes

def write_reg(addr, data):
    set_ui(data)
    set_uio_low(addr)           # wr_en=0, let wr_prev settle
    time.sleep_us(1)
    set_uio_high(addr | 0x08)   # wr_en=1, rising edge
    time.sleep_us(1)
    set_uio_low(addr)           # deassert

def absorb(byte):
    write_reg(1, byte)
    while read_busy():          # poll uio_out[0]
        pass

def squeeze():
    write_reg(0, 1)             # CMD squeeze
    result = []
    for i in range(11):
        result.append(read_uo_out())
        if i < 10:
            set_uio_high(2 | 0x10)  # rd_en=1, addr=2
            time.sleep_us(1)
            set_uio_low(0)
            time.sleep_us(1)
    return bytes(result)

# Hash b'\xAB\xCD\xEF' using hardware padding (CMD=2)
write_reg(0, 0)          # reset
absorb(0xAB)
absorb(0xCD)
absorb(0xEF)
write_reg(0, 2)          # hash: absorbs 0x81 pad byte and auto-squeezes
while read_busy():       # wait for pad permutation
    pass
digest = []
for i in range(11):
    digest.append(read_uo_out())
    if i < 10:
        set_uio_high(2 | 0x10)  # advance output
        time.sleep_us(1)
        set_uio_low(0)
        time.sleep_us(1)
print(bytes(digest).hex())

Using the cocotb simulation

cd test
pip install -r requirements.txt
make          # RTL simulation (iverilog + cocotb)
make WAVES=1  # also dump FST waveform (open with GTKWave or Surfer)

Nine test cases run automatically:

test_single_byte_absorb — absorbs 6 different single bytes, verifies digest
test_multi_byte_absorb — multi-byte sequences up to 11 bytes
test_absorb_timing — asserts exactly 25 cycles per absorb
test_out_valid_flag — checks out_valid transitions
test_reset_clears_state — same input after reset gives same digest
test_absorb_while_busy_ignored — writes during busy are silently dropped
test_reference_kat_components — validates Python model against published reference vectors (sBoxLayer and pLayer KATs, full LFSR sequence), then confirms DUT matches the validated model
test_vs_readable_crypto_reference — cross-checks against the independent joostrijneveld/readable-crypto implementation
test_hash_command — verifies CMD=2 applies pad 0x81 and auto-squeezes, matching hash88() from the reference model

Run the Python reference model standalone (no simulator needed):

cd test
python3 spongent88_ref.py

This prints the LFSR sequence, S-box checks, pLayer KAT, and digest values for standard inputs — useful for quickly catching spec mismatches before simulation.

Known-answer test vectors

From the BenchSpongent reference implementation and joostrijneveld/readable-crypto:

Input	Expected output
`sBoxLayer(0x0123456789ABCDEF012345)`	`0xEDB0214F7A859C36EDB021`
`pLayer(0x0123456789ABCDEF012345)`	`0x00FF003C3C333333155555`
LFSR[0..4]	`0x05, 0x0A, 0x14, 0x29, 0x13`

Hash KAT vectors (absorb single byte, no padding, squeeze full 88-bit state, LSB first):

Input byte	Digest (hex, 11 bytes)
`0x00`	`82f3cecf167feb3981c07c`
`0x01`	`0842dc1b6c7399eb92f540`
`0x80`	`a0623e32cd5a6bba0b304f`
`0xFF`	`fe511649a2fa375bf97aa3`
`0xA5`	`82b032622cbefe65b01911`

External hardware

No external hardware required. The chip is self-contained and communicates entirely through the standard TinyTapeout pin interface.

For W-OTS use, the host microcontroller (RP2040 on the demo board) handles:

Random private key generation (using its hardware RNG)
Key and signature storage (external flash or PSRAM recommended for full key sets)
W-OTS protocol logic (chain iteration, checksum, message formatting)
Padding bytes before the final absorb

IO

#	Input	Output	Bidirectional
0	data_in[0]	data_out[0]	busy
1	data_in[1]	data_out[1]	out_valid
2	data_in[2]	data_out[2]	addr[0]
3	data_in[3]	data_out[3]	addr[1] / wr_strobe
4	data_in[4]	data_out[4]	addr[2] / rd_strobe
5	data_in[5]	data_out[5]
6	data_in[6]	data_out[6]
7	data_in[7]	data_out[7]

106 Spongent-88 Hash Accelerator