diff --git a/.gitignore b/.gitignore
index 11614af..6299a37 100644
--- a/.gitignore
+++ b/.gitignore
@@ -112,4 +112,4 @@ venv.bak/
 dmypy.json
 
 # Pyre type checker
-.pyre/
+.pyre/
\ No newline at end of file
diff --git a/src/fastgm/der.py b/src/fastgm/der.py
new file mode 100644
index 0000000..eac870b
--- /dev/null
+++ b/src/fastgm/der.py
@@ -0,0 +1,301 @@
+import base64
+import warnings
+import binascii
+from itertools import chain
+from six import int2byte, b, text_type, integer_types
+
+_sentry = object()
+
+class FastgmExpectedDer(Exception):
+    pass
+
+# sm2 256 curve information
+sm256v1 = {
+    'ver': 1,
+    'sm2ecc_oid':[1, 2, 156, 10197, 1, 301],
+    'ecpk_oid':[1, 2, 840, 10045, 2, 1],
+    'header':'-----BEGIN EC PRIVATE KEY-----',
+}
+
+def str_idx_as_int(string, index):
+    """Take index'th byte from string, return as integer"""
+    val = string[index]
+    if isinstance(val, integer_types):
+        return val
+    return ord(val)
+
+def normalise_bytes(buffer_object):
+    """Cast the input into array of bytes."""
+    return memoryview(buffer_object).cast("B")
+
+def read_length(string):
+    if not string:
+        raise FastgmExpectedDer("Empty string can't encode valid length value")
+    num = str_idx_as_int(string, 0)
+    if not (num & 0x80):
+        # short form
+        return (num & 0x7F), 1
+    # else long-form: b0&0x7f is number of additional base256 length bytes,
+    # big-endian
+    llen = num & 0x7F
+    if not llen:
+        raise FastgmExpectedDer("Invalid length encoding, length of length is 0")
+    if llen > len(string) - 1:
+        raise FastgmExpectedDer("Length of length longer than provided buffer")
+    # verify that the encoding is minimal possible (DER requirement)
+    msb = str_idx_as_int(string, 1)
+    if not msb or llen == 1 and msb < 0x80:
+        raise FastgmExpectedDer("Not minimal encoding of length")
+    return int(binascii.hexlify(string[1 : 1 + llen]), 16), 1 + llen
+
+def remove_sequence(string):
+    if not string:
+        raise FastgmExpectedDer("Empty string does not encode a sequence")
+    if string[:1] != b"\x30":
+        n = str_idx_as_int(string, 0)
+        raise FastgmExpectedDer("wanted type 'sequence' (0x30), got 0x%02x" % n)
+    length, lengthlength = read_length(string[1:])
+    if length > len(string) - 1 - lengthlength:
+        raise FastgmExpectedDer("Length longer than the provided buffer")
+    endseq = 1 + lengthlength + length
+    return string[1 + lengthlength : endseq], string[endseq:]
+
+def remove_bitstring(string, expect_unused=_sentry):
+    if not string:
+        raise FastgmExpectedDer("Empty string does not encode a bitstring")
+    if expect_unused is _sentry:
+        warnings.warn(
+            "Legacy call convention used, expect_unused= needs to be"
+            " specified",
+            DeprecationWarning,
+        )
+    num = str_idx_as_int(string, 0)
+    if string[:1] != b"\x03":
+        raise FastgmExpectedDer("wanted bitstring (0x03), got 0x%02x" % num)
+    length, llen = read_length(string[1:])
+    if not length:
+        raise FastgmExpectedDer("Invalid length of bit string, can't be 0")
+    body = string[1 + llen : 1 + llen + length]
+    rest = string[1 + llen + length :]
+    if expect_unused is not _sentry:
+        unused = str_idx_as_int(body, 0)
+        if not 0 <= unused <= 7:
+            raise FastgmExpectedDer("Invalid encoding of unused bits")
+        if expect_unused is not None and expect_unused != unused:
+            raise FastgmExpectedDer("Unexpected number of unused bits")
+        body = body[1:]
+        if unused:
+            if not body:
+                raise FastgmExpectedDer("Invalid encoding of empty bit string")
+            last = str_idx_as_int(body, -1)
+            # verify that all the unused bits are set to zero (DER requirement)
+            if last & (2 ** unused - 1):
+                raise FastgmExpectedDer("Non zero padding bits in bit string")
+        if expect_unused is None:
+            body = (body, unused)
+    return body, rest
+
+def remove_integer(string):
+    if not string:
+        raise FastgmExpectedDer(
+            "Empty string is an invalid encoding of an integer"
+        )
+    if string[:1] != b"\x02":
+        n = str_idx_as_int(string, 0)
+        raise FastgmExpectedDer("wanted type 'integer' (0x02), got 0x%02x" % n)
+    length, llen = read_length(string[1:])
+    if length > len(string) - 1 - llen:
+        raise FastgmExpectedDer("Length longer than provided buffer")
+    if length == 0:
+        raise FastgmExpectedDer("0-byte long encoding of integer")
+    numberbytes = string[1 + llen : 1 + llen + length]
+    rest = string[1 + llen + length :]
+    msb = str_idx_as_int(numberbytes, 0)
+    if not msb < 0x80:
+        raise FastgmExpectedDer("Negative integers are not supported")
+    # check if the encoding is the minimal one (DER requirement)
+    if length > 1 and not msb:
+        # leading zero byte is allowed if the integer would have been
+        # considered a negative number otherwise
+        smsb = str_idx_as_int(numberbytes, 1)
+        if smsb < 0x80:
+            raise FastgmExpectedDer(
+                "Invalid encoding of integer, unnecessary "
+                "zero padding bytes"
+            )
+    return int(binascii.hexlify(numberbytes), 16), rest
+
+def remove_octet_string(string):
+    if string[:1] != b"\x04":
+        n = str_idx_as_int(string, 0)
+        raise FastgmExpectedDer("wanted type 'octetstring' (0x04), got 0x%02x" % n)
+    length, llen = read_length(string[1:])
+    body = string[1 + llen : 1 + llen + length]
+    rest = string[1 + llen + length :]
+    return body, rest
+
+def remove_ctx_t61string(string):
+    if not string:
+        raise FastgmExpectedDer("Empty string can't encode valid length value")
+    tag = str_idx_as_int(string, 0)
+    if (tag & 0xF0) != 0xA0:
+        raise FastgmExpectedDer("wanted type 'context-specify' (0xA0), got 0x%02x" % tag)
+    length, llen = read_length(string[1:])
+    if length > len(string) - 1 - llen:
+        raise FastgmExpectedDer("Length longer than the provided buffer")
+    endctx = 1 + llen + length
+    return string[1 + llen : endctx], string[endctx:]
+
+def encode_length(l):
+    assert l >= 0
+    if l < 0x80:
+        return int2byte(l)
+    s = ("%x" % l).encode()
+    if len(s) % 2:
+        s = b("0") + s
+    s = binascii.unhexlify(s)
+    llen = len(s)
+    return int2byte(0x80 | llen) + s
+
+def encode_number(n):
+    b128_digits = []
+    while n:
+        b128_digits.insert(0, (n & 0x7F) | 0x80)
+        n = n >> 7
+    if not b128_digits:
+        b128_digits.append(0)
+    b128_digits[-1] &= 0x7F
+    return b("").join([int2byte(d) for d in b128_digits])
+
+def encode_integer(r):
+    assert r >= 0  # can't support negative numbers yet
+    h = ("%x" % r).encode()
+    if len(h) % 2:
+        h = b("0") + h
+    s = binascii.unhexlify(h)
+    num = str_idx_as_int(s, 0)
+    if num <= 0x7F:
+        return b("\x02") + encode_length(len(s)) + s
+    else:
+        # DER integers are two's complement, so if the first byte is
+        # 0x80-0xff then we need an extra 0x00 byte to prevent it from
+        # looking negative.
+        return b("\x02") + encode_length(len(s) + 1) + b("\x00") + s
+
+def encode_bitstring(s, unused=_sentry):
+    encoded_unused = b""
+    len_extra = 0
+    if unused is _sentry:
+        warnings.warn(
+            "Legacy call convention used, unused= needs to be specified",
+            DeprecationWarning,
+        )
+    elif unused is not None:
+        if not 0 <= unused <= 7:
+            raise ValueError("unused must be integer between 0 and 7")
+        if unused:
+            if not s:
+                raise ValueError("unused is non-zero but s is empty")
+            last = str_idx_as_int(s, -1)
+            if last & (2 ** unused - 1):
+                raise ValueError("unused bits must be zeros in DER")
+        encoded_unused = int2byte(unused)
+        len_extra = 1
+    return b("\x03") + encode_length(len(s) + len_extra) + encoded_unused + s
+
+def encode_octet_string(s):
+    return b("\x04") + encode_length(len(s)) + s
+
+def encode_sequence(*encoded_pieces):
+    total_len = sum([len(p) for p in encoded_pieces])
+    return b("\x30") + encode_length(total_len) + b("").join(encoded_pieces)
+
+def encode_ctx_t61string(string, pos):
+    tag = 0xA0 + pos
+    return tag.to_bytes(1, 'big') + encode_length(len(string)) + string
+
+def encode_oid(first, second, *pieces):
+    assert 0 <= first < 2 and 0 <= second <= 39 or first == 2 and 0 <= second
+    body = b"".join(
+        chain(
+            [encode_number(40 * first + second)],
+            (encode_number(p) for p in pieces),
+        )
+    )
+    return b"\x06" + encode_length(len(body)) + body
+
+def unpem(pem):
+    if isinstance(pem, text_type):  # pragma: no branch
+        pem = pem.encode()
+
+    d = b("").join(
+        [
+            l.strip()
+            for l in pem.split(b("\n"))
+            if l and not l.startswith(b("-----"))
+        ]
+    )
+    return base64.b64decode(d)
+
+def to_pem(der, name):
+    b64 = base64.b64encode(der)
+    lines = [("-----BEGIN %s-----\n" % name).encode()]
+    lines.extend(
+        [b64[start : start + 64] + b("\n") for start in range(0, len(b64), 64)]
+    )
+    lines.append(("-----END %s-----\n" % name).encode())
+    return b("").join(lines)
+
+def sm2_pk_from_pem(pem):
+    unb64 = unpem(pem)
+    string = normalise_bytes(unb64)
+    total_seq, empty1 = remove_sequence(string)
+    oid_sequence, rest1 = remove_sequence(total_seq)
+    pk, rest2 = remove_bitstring(rest1, 0)
+    return pk[1:].hex()
+
+def sm2_sk_from_pem(pem):
+    pos = pem.find(sm256v1['header'])
+    unb64 = unpem(pem[pos:])
+    string = normalise_bytes(unb64)
+    total_seq, empty = remove_sequence(string)
+    version, ver_rest = remove_integer(total_seq)
+    sk, sk_rest = remove_octet_string(ver_rest)
+    sm2_ecc_ctx, sm2ecc_rest = remove_ctx_t61string(sk_rest)
+    pk_ctx, pk_ctx_rest = remove_ctx_t61string(sm2ecc_rest)
+    pk, rest = remove_bitstring(pk_ctx, 0)
+
+    return sk.hex(), pk[1:].hex()
+
+def sm2_pk_to_der(pbk):
+    pk = pbk.encode(encoding="utf-8")
+    pk_bitstring = encode_bitstring(b'\04' + pk, 0)
+    sm2ecc = encode_oid(sm256v1['sm2ecc_oid'][0], sm256v1['sm2ecc_oid'][1],\
+                        sm256v1['sm2ecc_oid'][2], sm256v1['sm2ecc_oid'][3],\
+                        sm256v1['sm2ecc_oid'][4], sm256v1['sm2ecc_oid'][5])
+    ecpk = encode_oid(sm256v1['ecpk_oid'][0], sm256v1['ecpk_oid'][1],\
+                      sm256v1['ecpk_oid'][2], sm256v1['ecpk_oid'][3],\
+                      sm256v1['ecpk_oid'][4], sm256v1['ecpk_oid'][5])
+    sequence = encode_sequence(ecpk, sm2ecc)
+    pk_der = encode_sequence(sequence, pk_bitstring)
+
+    return to_pem(pk_der, 'PUBLIC KEY')
+
+def sm2_sk_to_der(prk, pbk):
+    sk = prk.encode(encoding="utf-8")
+    pk = pbk.encode(encoding="utf-8")
+    # first parameter
+    sm2ecc = encode_oid(sm256v1['sm2ecc_oid'][0], sm256v1['sm2ecc_oid'][1],\
+                        sm256v1['sm2ecc_oid'][2], sm256v1['sm2ecc_oid'][3],\
+                        sm256v1['sm2ecc_oid'][4], sm256v1['sm2ecc_oid'][5])
+    # second parameter
+    pk_bitstring = encode_bitstring(b'\04' + pk, 0)
+    ctx_spec_1 = encode_ctx_t61string(pk_bitstring, 1)
+    ctx_spec_2 = encode_ctx_t61string(sm2ecc, 0)
+    sk_octet_string = encode_octet_string(sk)
+    version_integer = encode_integer(sm256v1['ver'])
+    sk_der = encode_sequence(version_integer, sk_octet_string, ctx_spec_2, ctx_spec_1)
+
+    cur_pem = to_pem(sm2ecc, 'EC PARAMETERS')
+    sk_pem  = to_pem(sk_der, 'EC PRIVATE KEY')
+    return cur_pem + sk_pem
diff --git a/src/fastgm/sm2.py b/src/fastgm/sm2.py
index f4bc319..7331fd1 100644
--- a/src/fastgm/sm2.py
+++ b/src/fastgm/sm2.py
@@ -1,6 +1,7 @@
 import os
 
 from .sm3 import hash, kdf
+from .der import sm2_pk_from_pem, sm2_sk_from_pem, sm2_pk_to_der, sm2_sk_to_der
 
 # 选择素域，设置椭圆曲线参数
 
@@ -221,13 +222,11 @@ def generate_key():
     
     return ('%064x'% k).upper(), point2hex(pk).upper()
 
-
 class SM2:
     def __init__(self, mode='C1C3C2'):
         """
         mode: C1C3C2 或 C1C2C3
         """
-        
         self._mode = mode
 
     @classmethod
@@ -251,4 +250,33 @@ def decrypt(self, sk, data):
         sk: 私钥, hex编码
         data: bytes
         """
-        return decrypt(sk, data, self._mode)
\ No newline at end of file
+        return decrypt(sk, data, self._mode)
+    
+    @classmethod
+    def load_pem(cls, pem,  is_sk = False):
+        '''
+        is_sk: True or False
+        is_sk = True
+        读取私钥pem, return (sk, pk)
+        is_sk = False
+        读取公钥pk, return pk
+        '''
+        if is_sk:
+            return sm2_sk_from_pem(pem)
+        else:
+            return sm2_pk_from_pem(pem)
+
+    @classmethod
+    def dump_pem(cls, sk, pk, is_sk = False):
+        '''
+        return sk-pem or pk-pem. 
+        '''
+        if is_sk:
+            if sk and pk: 
+                return sm2_sk_to_der(sk, pk)
+            else:
+                raise ValueError("Private key PEM need private key and public key simultaneously.")
+        else:
+            if pk == None:
+                raise ValueError("Must have public key.")
+            return sm2_pk_to_der(pk)
\ No newline at end of file
diff --git a/tests/test_gm.py b/tests/test_gm.py
index d7c38bf..e46e926 100644
--- a/tests/test_gm.py
+++ b/tests/test_gm.py
@@ -98,4 +98,37 @@ def test_sm2_generate_key():
     enc = sm2.encrypt(pk, data)
     dec = sm2.decrypt(sk, enc)
 
-    assert dec == data
\ No newline at end of file
+    assert dec == data
+
+def test_sm2_form_pem():
+    '''
+    使用openssl 生成SM2密钥
+    openssl ecparam -genkey -name SM2 -out sk.pem
+    openssl ec -in sk.pem -pubout -out pk.pem
+    '''
+    with open('sk.pem') as f:
+        sk_buffer = f.read()
+
+    with open('pk.pem') as f:
+        pk_buffer = f.read()
+
+    sk, pk1 = SM2.load_pem(sk_buffer, True)
+    pk = SM2.load_pem(pk_buffer, False)
+    print(sk)
+    print(pk)
+
+def test_sm2_sk_to_pem():
+    '''
+    将生成的密钥保存到pem
+    pem格式符合OpenSSL
+    '''
+    sk, pk = SM2.generate_key()
+
+    with open('test_sk.pem', "wb") as f:
+        f.write(SM2.dump_pem(sk, pk, True))
+
+def test_sm2_pk_to_pem():
+    sk, pk = SM2.generate_key()
+
+    with open('test_pk.pem', "wb") as f:
+        f.write(SM2.dump_pem(None, pk, False))
\ No newline at end of file