diff --git a/wolfcrypt/src/dilithium.c b/wolfcrypt/src/dilithium.c
index 4bf1cdd223..d3158dc50c 100644
--- a/wolfcrypt/src/dilithium.c
+++ b/wolfcrypt/src/dilithium.c
@@ -60,7 +60,11 @@
  *   is considerably slower.
  * WOLFSSL_DILITHIUM_SIGN_SMALL_MEM_PRECALC                   Default: OFF
  *   Compiles signature implementation that uses smaller amounts of memory but
- *   is considerably slower. Allocates vectors for decoded parameters.
+ *   is considerably slower. Allocates vectors and decodes private key data
+ *   into them upfront.
+ * WOLFSSL_DILITHIUM_SIGN_SMALL_MEM_PRECALC_A                 Default: OFF
+ *   Compiles signature implementation that uses smaller amounts of memory but
+ *   is slower. Allocates matrix A and calculates it upfront.
  *
  * WOLFSSL_DILITHIUM_ALIGNMENT                                Default: 8
  *   Use to indicate whether loading and storing of words needs to be aligned.
@@ -145,6 +149,13 @@
         !defined(WOLFSSL_DILITHIUM_SIGN_SMALL_MEM)
     #define WOLFSSL_DILITHIUM_SIGN_SMALL_MEM
 #endif
+#if defined(WOLFSSL_DILITHIUM_SIGN_SMALL_MEM_PRECALC_A) && \
+        !defined(WOLFSSL_DILITHIUM_SIGN_SMALL_MEM)
+    #define WOLFSSL_DILITHIUM_SIGN_SMALL_MEM
+    #ifdef WOLFSSL_DILITHIUM_SIGN_SMALL_MEM_PRECALC
+        #error "PRECALC and PRECALC_A is equivalent to non small mem"
+    #endif
+#endif
 
 #ifdef WOLFSSL_WC_DILITHIUM
 
@@ -2742,7 +2753,8 @@ static void dilithium_decompose_q32(sword32 r, sword32* r0, sword32* r1)
 
 #ifndef WOLFSSL_DILITHIUM_NO_SIGN
 
-#ifndef WOLFSSL_DILITHIUM_SIGN_SMALL_MEM
+#if !defined(WOLFSSL_DILITHIUM_SIGN_SMALL_MEM) || \
+    defined(WOLFSSL_DILITHIUM_SIGN_SMALL_MEM_PRECALC_A)
 /* Decompose vector of polynomials into high and low based on GAMMA2.
  *
  * @param [in]  r       Vector of polynomials to decompose.
@@ -5766,9 +5778,11 @@ static int dilithium_sign_msg_with_seed(dilithium_key* key, const byte* seed,
         /* y-l, w0-k, w1-k, c-1, z-1, A-1 */
         allocSz  = params->s1Sz + params->s2Sz + params->s2Sz +
             DILITHIUM_POLY_SIZE +  DILITHIUM_POLY_SIZE + DILITHIUM_POLY_SIZE;
-#ifdef WOLFSSL_DILITHIUM_SIGN_SMALL_MEM_PRECALC
+    #ifdef WOLFSSL_DILITHIUM_SIGN_SMALL_MEM_PRECALC
         allocSz += params->s1Sz + params->s2Sz + params->s2Sz;
-#endif
+    #elif defined(WOLFSSL_DILITHIUM_SIGN_SMALL_MEM_PRECALC_A)
+        allocSz += params->aSz - DILITHIUM_POLY_SIZE;
+    #endif
         y = (sword32*)XMALLOC(allocSz, NULL, DYNAMIC_TYPE_DILITHIUM);
         if (y == NULL) {
             ret = MEMORY_E;
@@ -5780,7 +5794,11 @@ static int dilithium_sign_msg_with_seed(dilithium_key* key, const byte* seed,
             z     = c  + DILITHIUM_N;
             a     = z  + DILITHIUM_N;
             ct0   = z;
+        #ifndef WOLFSSL_DILITHIUM_SIGN_SMALL_MEM_PRECALC_A
             y_ntt = z;
+        #else
+            y_ntt = w0;
+        #endif
         #ifndef WOLFSSL_DILITHIUM_SIGN_SMALL_MEM_PRECALC
             s1    = z;
             s2    = z;
@@ -5811,6 +5829,13 @@ static int dilithium_sign_msg_with_seed(dilithium_key* key, const byte* seed,
     if (ret == 0) {
         dilithium_make_priv_vecs(key, s1, s2, t0);
     }
+#endif
+#ifdef WOLFSSL_DILITHIUM_SIGN_SMALL_MEM_PRECALC_A
+    if (ret == 0) {
+        /* Step 5: Create the matrix A from the public seed. */
+        ret = dilithium_expand_a(&key->shake, pub_seed, params->k,
+            params->l, a);
+    }
 #endif
     if (ret == 0) {
         word16 kappa = 0;
@@ -5823,12 +5848,14 @@ static int dilithium_sign_msg_with_seed(dilithium_key* key, const byte* seed,
             byte* commit = sig;
             byte r;
             byte s;
-            byte aseed[DILITHIUM_GEN_A_SEED_SZ];
             sword32 hi;
-            sword32* at = a;
-            sword32* wt = w;
             sword32* w0t = w0;
             sword32* w1t = w1;
+        #ifndef WOLFSSL_DILITHIUM_SIGN_SMALL_MEM_PRECALC_A
+            byte aseed[DILITHIUM_GEN_A_SEED_SZ];
+            sword32* at = a;
+            sword32* wt = w;
+        #endif
 
             valid = 1;
             /* Step 12: Compute vector y from private random seed and kappa. */
@@ -5839,6 +5866,7 @@ static int dilithium_sign_msg_with_seed(dilithium_key* key, const byte* seed,
                 (1 << params->gamma1_bits) - params->beta);
         #endif
 
+        #ifndef WOLFSSL_DILITHIUM_SIGN_SMALL_MEM_PRECALC_A
             /* Step 5: Create the matrix A from the public seed. */
             /* Copy the seed into a buffer that has space for s and r. */
             XMEMCPY(aseed, pub_seed, DILITHIUM_PUB_SEED_SZ);
@@ -5906,6 +5934,16 @@ static int dilithium_sign_msg_with_seed(dilithium_key* key, const byte* seed,
                 w0t += DILITHIUM_N;
                 w1t += DILITHIUM_N;
             }
+        #else
+            /* Step 13: NTT-1(A o NTT(y)) */
+            XMEMCPY(y_ntt, y, params->s1Sz);
+            dilithium_vec_ntt(y_ntt, params->l);
+            dilithium_matrix_mul(w, a, y_ntt, params->k, params->l);
+            dilithium_vec_invntt(w, params->k);
+            /* Step 14, Step 22: Make values positive and decompose. */
+            dilithium_vec_make_pos(w, params->k);
+            dilithium_vec_decompose(w, params->k, params->gamma2, w0, w1);
+        #endif
             if ((ret == 0) && valid) {
                 sword32* yt = y;
             #ifndef WOLFSSL_DILITHIUM_SIGN_SMALL_MEM_PRECALC