If you want to develop a new POW coin, you can't use SHA256 directly, but you can use my method, because the existing ASIC machine is not good at process massive memory operation.
Again look into yescrypt for example. Memory hard algorithms aren't new. Even scrypt can be asic resistance with high enough N value set.
How is yours different than these?
My algorithm combine the advantages of both SHA256 and memory access, SHA256 have been proved in bitcoin for many years.
So basically what your saying is your algorithm can be a plug in replacement for sha256? So you could plug this directly into bitcoin without any major code changes? Soft fork, hard fork?
Here is Yescrypt pseudocode:
# ** Functions/symbols **
# || Concatenate two strings
# HMAC(h, k, v) HMAC with hash function h and key k over value v
# PBKDF2(prf, p, s, c, dklen) PBKDF2 key derivation function
# substr(s, start, length) Substring from start (zero-based) of length bytes
# le32dec(), le32enc() 32-bit little-endian decoding/encoding
# SIMD_[un]shuffle() Salsa20 SIMD (un)shuffling of 32-bit words
# Integerify(B, r) Parse B_{2r-1} as a little-endian integer
# p2floor(x) Largest power of 2 not greater than x
# ** Inputs **
string password
string salt
integer t_cost
integer m_cost
integer outlen
# ** Algorithm **
N = 8 << m_cost
r = 8
# ** Settings hard-coded/assumed below in this pseudocode **
# p = 1
# g = 0
# flags = YESCRYPT_RW
# no ROM
# If m_cost is 16 MB per thread or more, pre-hash using 1/64th of m_cost first,
# to mitigate garbage collector attacks. yescrypt_prehash() is almost the same
# as this function, but its personalization HMAC key is "yescrypt-prehash"
# rather than "yescrypt", it skips builtin SCRAM finalization, and it will not
# invoke another yescrypt_prehash().
if (N / p >= 0x100 && N / p * r >= 0x20000)
password = yescrypt_prehash(password, salt, t_cost, m_cost / 64, 32)
password = HMAC(SHA-256, "yescrypt", password)
B = PBKDF2(HMAC-SHA-256, password, salt, 1, 128 * r)
password = substr(B, 0, 32)
# SMix1 invoked to initialize pwxform S-boxes
X = SIMD_shuffle(le32dec(B))
for i = 0 to Sbytes/128 - 1
S[i] = X
X = BlockMix_{Salsa20/8, 1}(X)
# SMix1 invoked "for real"
for i = 0 to N - 1
V[i] = X
if (i > 1)
j = Wrap(Integerify(X, r), i)
X = X xor V[j]
X = BlockMix_pwxform{Salsa20/2, S, r}(X)
# SMix2
if (t_cost = 0)
Nloop = (N + 2) / 3
else if (t_cost = 1)
Nloop = (N * 2 + 2) / 3
else
Nloop = N * (t - 1)
for i = 0 to Nloop - 1
j = Integerify(X, r) mod N
X = X xor V[j]
V[j] = X
X = BlockMix_pwxform{Salsa20/2, S, r}(X)
B = le32enc(SIMD_unshuffle(X))
out = PBKDF2(HMAC-SHA-256, password, B, 1, outlen)
# Builtin SCRAM (RFC 5802) support
clen = min(outlen, 32)
dk = PBKDF2(HMAC-SHA-256, password, B, 1, 32)
dk = SHA-256(HMAC(SHA-256, dk, "Client Key"))
out = substr(dk, 0, clen) || substr(out, clen, outlen - clen)
return out
# ** Helper functions **
# Wrap x to the range 0 to i-1
Wrap(x, i)
n = p2floor(i)
return (x mod n) + (i - n)
