You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Open hardware-based and patent-free Physical Unclonable Function (PUF) #3064
PUF technology is an emerging technology to store the essential part of a self-sovereign identity, the private key, in a tamper-proof manner. The private key is storage becomes volatile by using PUF technology [Patents].
More sophisticated 2015 work
PUFs in general, SRAM PUFs in particular, can be used as a secure cryptographic key storage mechanism . Fig. 1 shows how such mechanism can be integrated to create a PUF based key storage system.
The popular 6Ts SRAM cell (see Fig. 2(a)) consists of two cross-coupled CMOS inverters formed by four transistors (Q 1 with Q5 and Q2 with Q6) and two pass transistors (Q3 and Q4). The pass transistors are used to access the cell for read and write operations. The bitline (BL), the compliment bitline (BLB) and the wordline (WL) are used to access the cell.
We simulate the start-up behavior of an SRAM cell using SPICE and BSIM4 65nm models. Process variation PDF for 65nm :
physical part: a physical system that is very difficult to clone due to uncontrollable process variations during manufacturing
operational part: a set of challenges Ci(stimuli) has to be available to which the system responds with a set of sufficiently different responses Ri
Ri <- PUF(Ci) : a response Ri is generated when challenging a PUF with a challenge Ci
challenge: input to PUF
response: output from PUF
challenge response pair / CRP: an applied challenged and its measured response
Two distance measures of PUFs
inter distance: distance between applying a challenge to two different PUFs
intra distance: distance between applying a challenge twice to a PUF
Type of PUFs based on design
non electronic PUFs
optical PUF
paper PUF
CD PUF
RF-DNA
magnetic PUF
acoustic PUF
analog electronic PUFs
VT PUF
power distribution PUF
coating PUF
LC PUF
intrinsic PUFs
requirement to be intrinsic
PUF and the measurement equipment should be fully integrated
has to be available naturally during the manufacturing process
delay based intrinsic PUFs
arbiter PUF: introduce a digital race condition on two paths on a chip and have an arbiter circuit to decide which one won the race
attack on arbiter PUFs
model building attack: attack using a mathematical model of the PUF created from observing CRP queries which able to predict the response to an unseen challenge with relatively high accuracy
ring oscillator PUF: output of a digital line is inverted and fed back to its input, creating an asynchronously oscillating loop.
attack:
driving a sinusoidal signal on the ground plane of a ring oscillator can cause it to lock to that signal
electromagnetic radiation from the ring oscillator PUF can be used to steal the output bits
memory based intrinsic PUFs
SRAM PUF: random physical mismatch in the cell caused by manufacturing variability determines the power up behavior (can be zero, one, or no preference).
need a device powered up to enable response generator
if a SRAM cell is powered on and set to the '0' for a long time(~10 days), then on subsequent power-on sequences, the cell is more likely to skew toward the '1' state.
attack:
if one could insert a write command, then it could leverage the NBTI(negative bias temperature instability) to deliberately force individual bits toward 1
if one could modify the temperature, one could potentially cause the PUF to fail by running the PUF outside its design area
an attacker with access to the power channel could potentially control the stability of some of the SRAM PUF bits
butterfly PUF: cross coupling two transparent data latches. using the clear functionalities of the latches, an unstable state can be introduced after which the circuit converges back to one of the two stable states.
introduced because SRAM PUF is not possible in FPGA since the SRAM cells is always reseted to zero after power up in FPGA
doesn't need a device powered up. can be run using a reset button
latch PUF: two NOR gates are cross coupled. it will converge to a stable state depending on the internal mismatch between the electronic components.
flip flop PUF
inter and intra distance can be improved using 1-out-of-9 majority voting
conceptual PUFs
physically obfuscated keys(POK): a key is permanently stored in a physical way, makes it hard to learn the key by probing attack.
invasive attack should destroy the key and make further use impossible
controlled PUF(CPUF): combination of PUF and a control layer in which the PUF is inseparably embedded.
advantage:
a hash function to generate the responses of the PUF can prevent chosen challenge attacks
final responses more reliable because exists an error correction algorithm
no link between the responses and the physical details because the error corrected outputs is already affected by the hash function
the hash function generating the challenges can take additional inputs
reconfigurable PUF
extends the regular CRP behavior of a PUF with an additional operation(reconfiguration)
quantum readout PUF
replace the regular challenges and responses with qunatum states
SIMPL system and PPUFs: use PUF as a part of a public key like algorithm. rely on systems which can be modeled
SIMPL: simulation possible but laborious
PPUF: public PUF
Type of PUFs based on number CRPs
strong PUFs
properties
large enough CRPs
stable responses
unable to be predicted on a new challenge
not feasible to manufacture two PUFs with the same response
readout only reveals the response, no internal functionality is leaked
typically used for authentication
ex: non electronic PUFs
weak PUFs
properties:
a small number of CRPs
responses are stable and robust to various conditions and multiple readings
responses are unpredictable
impractical to manufacture two devices with the same physical fingerprint
usually used for key storage
ex: intrinsic PUFs
Property of PUF
evaluatable
unique
reproducible
unclonable
unpredictable
one way
tamper evident
Technique to reduce noise
use differential design techniques to cancel out first order environmental dependencies
employ multiple error correction techniques. may leak bits of the secret key
use soft decision coding: take advantage of the reliability information of a given response bit to improve error correction performance
Application
authentication using strong PUFs
phase
enrollment: gather CRPs from a PUF then stored its measured response. server must either store enough CRPs so that it will not run out, or it must periodically recharge the table by establishing secure communication with an authenticated client and requesting responses to new challenges.
verification: apply a challenge from CRP database and compare the PUF response with the corresponding response from the database
cryptographic key generation using weak PUFs
using SRAM: power on the SRAM and observe the memory state
using ring oscillator PUF: pairwise compares each of the oscillators in order to measure the correct ordering of oscillation frequency
statistical and systematic noise must be corrected/mitigated to get a stable set of unique bits
secret key generation using two phase algorithm
in generation phase, the PUF is queried and the algorithm produces a secret key together with some additional information called helper data. both stored in a db by the verifier
in reproduction phase, the verifier presents the helper data to the algorithm which uses it to extract the same key from the PUF as in the generation step. thus, the PUF and the verifier have established a shared secret key
authentication using weak PUFs
get unique bits using steps same as key generation. supplement the weak PUF with a hardware HMAC(hash message authentication code) or AES implementation.
hardware entangled cryptography
keyless: never stored a secret digital key in the memory during running the algorithm
We have various Raspberry Pi model B+ 1.2 hardware. This enables quick and dirty prototyping, possibly even using the Python language :-)
Goal: obtain SRAM chips and build the first completely patent-free functional PUF. Entropy and efficiency are not yet important. Key goal is raw uninitialized memory readout:
base file for comparison: ascii-coded binary file 171017B/1.c
Algorithm for PUF using repetition code
generate/read key (128 bits)
produce a repetition of the key (here, repetition length is 16, 64 or 128) (128 bits x 128/64/16 repetition = 16k / 8k / 2k bits)
first bits (16k / 8k / 2k bits) of the base file are XORed with repetition of bit key to produce helper data.
helper data is XOR with bit from the destination file.
decode repetition encoding (vote the majority of the result of previous step to reconstruct the key).
compare the reconstructed key with the actual key
repetition: 128
Chip
Folder
File
Difference
Difference (%)
Chip
Folder
File
Difference
Difference (%)
A
171017A
1
6
4.7%
B
171017B
1
0
0.0%
2
0
0.0%
2
0
0.0%
3
11
8.6%
3
0
0.0%
4
1
0.8%
4
0
0.0%
5
8
6.3%
5
0
0.0%
6
3
2.3%
6
0
0.0%
7
3
2.3%
7
0
0.0%
8
3
2.3%
8
0
0.0%
9
1
0.8%
9
0
0.0%
10
1
0.8%
10
0
0.0%
C
171017C
1
0
0.0%
D
171017D
1
1
0.8%
2
0
0.0%
2
0
0.0%
3
0
0.0%
3
4
3.1%
4
0
0.0%
4
0
0.0%
5
7
5.5%
5
8
6.3%
6
1
0.8%
6
0
0.0%
7
0
0.0%
7
0
0.0%
8
1
0.8%
8
0
0.0%
9
0
0.0%
9
0
0.0%
10
3
2.3%
10
0
0.0%
E
171017E
1
1
0.8%
F
181017F
1
58
45.3%
2
0
0.0%
2
32
25.0%
3
9
7.0%
3
45
35.2%
4
2
1.6%
4
37
28.9%
5
4
3.1%
5
48
37.5%
6
1
0.8%
6
57
44.5%
7
0
0.0%
7
77
60.2%
8
0
0.0%
8
82
64.1%
9
0
0.0%
9
85
66.4%
10
20
15.6%
10
94
73.4%
G
181017G
1
76
59.4%
H
181017H
1
30
23.4%
2
18
14.1%
2
13
10.2%
3
28
21.9%
3
34
26.6%
4
29
22.7%
4
30
23.4%
5
9
7.0%
5
12
9.4%
6
21
16.4%
6
17
13.3%
7
41
32.0%
7
26
20.3%
8
42
32.8%
8
22
17.2%
9
12
9.4%
9
18
14.1%
10
34
26.6%
10
13
10.2%
I
181017I
1
11
8.6%
J
181017J
1
18
14.1%
2
0
0.0%
2
82
64.1%
3
4
3.1%
3
70
54.7%
4
0
0.0%
4
31
24.2%
5
0
0.0%
5
64
50.0%
6
0
0.0%
6
60
46.9%
7
0
0.0%
7
107
83.6%
8
1
0.8%
8
69
53.9%
9
0
0.0%
9
23
18.0%
10
9
7.0%
10
44
34.4%
repetition: 64
Chip
Folder
File
Difference
Difference (%)
Chip
Folder
File
Difference
Difference (%)
A
171017A
1
2
1.6%
B
171017B
1
0
0.0%
2
5
3.9%
2
0
0.0%
3
13
10.2%
3
0
0.0%
4
0
0.0%
4
0
0.0%
5
10
7.8%
5
0
0.0%
6
16
12.5%
6
0
0.0%
7
0
0.0%
7
0
0.0%
8
0
0.0%
8
0
0.0%
9
9
7.0%
9
0
0.0%
10
1
0.8%
10
0
0.0%
C
171017C
1
8
6.3%
D
171017D
1
7
5.5%
2
6
4.7%
2
1
0.8%
3
17
13.3%
3
12
9.4%
4
1
0.8%
4
3
2.3%
5
29
22.7%
5
13
10.2%
6
8
6.3%
6
1
0.8%
7
15
11.7%
7
1
0.8%
8
9
7.0%
8
1
0.8%
9
6
4.7%
9
1
0.8%
10
17
13.3%
10
7
5.5%
E
171017E
1
17
13.3%
F
181017F
1
71
55.5%
2
20
15.6%
2
39
30.5%
3
32
25.0%
3
49
38.3%
4
21
16.4%
4
35
27.3%
5
22
17.2%
5
51
39.8%
6
24
18.8%
6
63
49.2%
7
18
14.1%
7
105
82.0%
8
25
19.5%
8
108
84.4%
9
18
14.1%
9
128
100.0%
10
37
28.9%
10
128
100.0%
G
181017G
1
66
51.6%
H
181017H
1
35
27.3%
2
51
39.8%
2
21
16.4%
3
48
37.5%
3
34
26.6%
4
52
40.6%
4
28
21.9%
5
50
39.1%
5
24
18.8%
6
68
53.1%
6
10
7.8%
7
57
44.5%
7
34
26.6%
8
45
35.2%
8
27
21.1%
9
59
46.1%
9
20
15.6%
10
52
40.6%
10
23
18.0%
I
181017I
1
11
8.6%
J
181017J
1
18
14.1%
2
7
5.5%
2
119
93.0%
3
8
6.3%
3
18
14.1%
4
6
4.7%
4
48
37.5%
5
5
3.9%
5
22
17.2%
6
8
6.3%
6
17
13.3%
7
7
5.5%
7
128
100.0%
8
7
5.5%
8
28
21.9%
9
7
5.5%
9
49
38.3%
10
9
7.0%
10
18
14.1%
repetition: 16
Chip
Folder
File
Difference
Difference (%)
Chip
Folder
File
Difference
Difference (%)
A
171017A
1
27
21.1%
B
171017B
1
0
0.0%
2
15
11.7%
2
0
0.0%
3
33
25.8%
3
0
0.0%
4
22
17.2%
4
0
0.0%
5
30
23.4%
5
0
0.0%
6
27
21.1%
6
0
0.0%
7
14
10.9%
7
0
0.0%
8
27
21.1%
8
0
0.0%
9
21
16.4%
9
0
0.0%
10
21
16.4%
10
0
0.0%
C
171017C
1
48
37.5%
D
171017D
1
35
27.3%
2
52
40.6%
2
34
26.6%
3
44
34.4%
3
33
25.8%
4
50
39.1%
4
38
29.7%
5
37
28.9%
5
34
26.6%
6
50
39.1%
6
42
32.8%
7
57
44.5%
7
39
30.5%
8
47
36.7%
8
34
26.6%
9
48
37.5%
9
33
25.8%
10
49
38.3%
10
36
28.1%
E
171017E
1
28
21.9%
F
181017F
1
35
27.3%
2
40
31.3%
2
120
93.8%
3
31
24.2%
3
26
20.3%
4
42
32.8%
4
27
21.1%
5
33
25.8%
5
29
22.7%
6
43
33.6%
6
26
20.3%
7
38
29.7%
7
40
31.3%
8
29
22.7%
8
120
93.8%
9
38
29.7%
9
120
93.8%
10
31
24.2%
10
120
93.8%
G
181017G
1
65
50.8%
H
181017H
1
17
13.3%
2
58
45.3%
2
15
11.7%
3
56
43.8%
3
16
12.5%
4
54
42.2%
4
25
19.5%
5
52
40.6%
5
19
14.8%
6
61
47.7%
6
27
21.1%
7
74
57.8%
7
20
15.6%
8
54
42.2%
8
19
14.8%
9
69
53.9%
9
25
19.5%
10
60
46.9%
10
12
9.4%
I
181017I
1
19
14.8%
J
181017J
1
35
27.3%
2
24
18.8%
2
35
27.3%
3
11
8.6%
3
120
93.8%
4
7
5.5%
4
35
27.3%
5
16
12.5%
5
111
86.7%
6
15
11.7%
6
40
31.3%
7
22
17.2%
7
35
27.3%
8
21
16.4%
8
120
93.8%
9
14
10.9%
9
44
34.4%
10
30
23.4%
10
113
88.3%
Result:
0% error should only produced when comparing with the chip used as the base file
16 as repetition code is acceptable
Choosing too large repetition code makes multiple chips resulting in the same key. This is undesirable. Thus, choosing a correct length that can correct acceptable error rate in the same chip without introducing two correct key from different chips is a challenge.
Further improvement:
use a better ECC
choose a stable bit as the input for XOR (here, we only choose the first "length of repetition * length of key" bits)
Please provide @qstokkink with 4 out of your 5 operational SRAM puf chips. Three Arduino + interface board plus sources to get it operational and profile each PUF bit. Allows us to do parallel profiling. Additional microSDcard for storage of helper data + adapter to interface with Arduino.
Figure 4.8 + 4.9 Remanence Graph of CY62256NLL; please provide raw measurements in graph. Optionally connect those sample points with a line, remove smooth polished line.
Discuss your code coverage and provide a table with lines of code + percentage covered.
sell your stuff even more, like this in abstract:
As the grand concluding experiment of this thesis we store the private key of a Bitcoin with the physical security protection of our PUF.
@synctext I think the issue title should be changed into "Open software-based and patent-free Physical Unclonable Function (PUF) " not "Open hardware-based and patent-free Physical Unclonable Function (PUF) "
What makes this patent free?
The technologies as described have numerous patents that cover the SRAM-PUF mechanisms.
@nymble Good point. We have not yet done a thorough prior work. This thesis is build upon early scientific papers and early work by Delft ourselves. The 2003 patents describe systems like "During an enrollment phase, Alice 16 issues a challenge C to source P 20 and receives a response A from source P 20". Our approach avoids this with "store this given private key generation seed".
how the generated secret key and seed for the random number are stored inside the secure storage of sram?
Is the secret key ever stored inside in the sram or it is generated everytime we need to encrypt or decrypt anything?
@Rahulsharma311 This is a stand-alone implementation for research purposes.
Is the secret key ever stored inside in the sram
The secret key is generated from physical properties upon startup. As stated above, "the private key is storage becomes volatile by using PUF". There is no secure sram storage, the PUF is designed to be airgapped and challenge/response interfaced as an isolated component. In a decade this could become a standard component within a smartphone to achieve passport-grade digital identity.
New master students always welcome.
This project has been without an active master student for a while. Closing this issue until a new person can work on this.
I am working on my thesis "MAchine learning attacks on DRAM PUF" as part of my thesis, am currently looking for a dataset If anyone has a DRAM PUF dataset it would be really helpful.
Activity
asajim commentedon Aug 22, 2017
synctext commentedon Aug 22, 2017
using standard PC board, "Investigating SRAM PUFs in large CPUs and GPUs" with code listings: https://arxiv.org/pdf/1507.08514.pdf
DDR3: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4168438/
https://users.wpi.edu/~martin/MQP/edwardsetal.pdf
asajim commentedon Sep 5, 2017
Note:
2 main parts of PUF
Ri <- PUF(Ci) : a response Ri is generated when challenging a PUF with a challenge Ci
Two distance measures of PUFs
Type of PUFs based on design
Type of PUFs based on number CRPs
Property of PUF
Technique to reduce noise
Application
asajim commentedon Sep 5, 2017
Extra readings:
synctext commentedon Sep 6, 2017
We have various Raspberry Pi model B+ 1.2 hardware. This enables quick and dirty prototyping, possibly even using the Python language :-)



Goal: obtain SRAM chips and build the first completely patent-free functional PUF. Entropy and efficiency are not yet important. Key goal is raw uninitialized memory readout:
synctext commentedon Sep 21, 2017
First operational prototype! @hakhundov thnx for the help

Solid progress within this master thesis project, it only started a few weeks ago.
Hardware: Arduino microcontroller board based on the ATmega2560 and the attached 1 Mbit Serial SRAM device, a 23LC1024 by Microchip
The skeleton open source repo for dumping the raw uninitialized memory using SPI. Next steps:
synctext commentedon Oct 10, 2017
Possibly focus on the unstable bit plots of your hardware (see related work with 1432 repeats in their experiment).
asajim commentedon Oct 10, 2017
Screenshot of some initialized bit from SRAM 23LC1024

asajim commentedon Oct 19, 2017
Analysis result of 10 SRAMs 23LC1024:
Length of bits: 524288
Error rate : HDintra
Chip A
Chip B
Chip C
Chip D
Chip E
Chip F
Chip G
Chip H
Chip I
Chip J
Correlation between chips : HDinter
Symmetry / Mean Value : μ
Value on x-axis: 0 means stable 0, 10 means stable 1 (during 10 profiling trials always show value 1).
Chip A

Chip B

Chip C

Chip D

Chip E

Chip F

Chip G

Chip H

Chip I

Chip J

Correlation between bits
Chip A

Chip B

Chip C

Chip D

Chip E

Chip F

Chip G

Chip H

Chip I

Chip J

Summary
asajim commentedon Oct 21, 2017
Analysis result of 10 SRAMs 23K256:
Error rate : HDintra
Chip A
Chip B
Chip C
Chip D
Chip E
Chip F
Chip G
Chip H
Chip I
Correlation between chips : HDinter
Symmetry / Mean Value : μ
Value on x-axis: 0 means stable 0, 10 means stable 1 (during 10 profiling trials always show value 1).
Chip A

Chip B

Chip C

Chip D

Chip E

Chip F

Chip G

Chip H

Chip I

Correlation between bits
Chip A

Chip B

Chip C

Chip D

Chip E

Chip F

Chip G

Chip H

Chip I

Summary
* 23K256 show more unreliable result compared to 23LC1024. For example, the error rate is higher than 23LC1024.asajim commentedon Oct 23, 2017
Result of PUF trial using repetition code as ECC
key length: 128 bitkey:
01010111001100110100000101010001
00101011011110000110110001010001
01110011010000110100011101101011
01000001011001110011110100111101
base file for comparison: ascii-coded binary file 171017B/1.c
Algorithm for PUF using repetition code
repetition: 128
repetition: 64
repetition: 16
Result:
27 remaining items
asajim commentedon Apr 20, 2018
Thesis draft:
thesis.pdf
synctext commentedon Apr 20, 2018
Please provide @qstokkink with 4 out of your 5 operational SRAM puf chips. Three Arduino + interface board plus sources to get it operational and profile each PUF bit. Allows us to do parallel profiling. Additional microSDcard for storage of helper data + adapter to interface with Arduino.
synctext commentedon Apr 20, 2018
asajim commentedon May 3, 2018
Latest draft:
thesis.pdf
synctext commentedon May 7, 2018
sell your stuff even more, like this in abstract:
As the grand concluding experiment of this thesis we store the private key of a Bitcoin with the physical security protection of our PUF.
asajim commentedon May 10, 2018
Final thesis report: thesis.pdf
Thesis repository : software-based-PUF
@synctext I think the issue title should be changed into "Open software-based and patent-free Physical Unclonable Function (PUF) " not "Open hardware-based and patent-free Physical Unclonable Function (PUF) "
synctext commentedon May 18, 2018
Congrats on your TUDelft master degree! Official thesis .pdf download repo: https://repository.tudelft.nl/islandora/object/uuid%3A4f879ecf-95d5-4482-8931-8c40abde0e79
"Open-Source Software-Based SRAM-PUF for Secure Data and Key Storage Using Off-The-Shelf SRAM"
nymble commentedon Sep 4, 2018
What makes this patent free?
The technologies as described have numerous patents that cover the SRAM-PUF mechanisms.
synctext commentedon Nov 3, 2018
@nymble Good point. We have not yet done a thorough prior work. This thesis is build upon early scientific papers and early work by Delft ourselves. The 2003 patents describe systems like "During an enrollment phase, Alice 16 issues a challenge C to source P 20 and receives a response A from source P 20". Our approach avoids this with "store this given private key generation seed".
manasviaudichya311 commentedon Dec 5, 2019
how the generated secret key and seed for the random number are stored inside the secure storage of sram?
Is the secret key ever stored inside in the sram or it is generated everytime we need to encrypt or decrypt anything?
synctext commentedon Dec 12, 2019
@Rahulsharma311 This is a stand-alone implementation for research purposes.
The secret key is generated from physical properties upon startup. As stated above, "the private key is storage becomes volatile by using PUF". There is no secure sram storage, the PUF is designed to be airgapped and challenge/response interfaced as an isolated component. In a decade this could become a standard component within a smartphone to achieve passport-grade digital identity.
synctext commentedon Jul 20, 2020
New master students always welcome.
This project has been without an active master student for a while. Closing this issue until a new person can work on this.
reshmisuragani commentedon Mar 5, 2021
Hello,
I am working on my thesis "MAchine learning attacks on DRAM PUF" as part of my thesis, am currently looking for a dataset If anyone has a DRAM PUF dataset it would be really helpful.
synctext commentedon Mar 5, 2021
interesting, machine learning attacks..
Have you seen the data above 'result of 10 SRAMs 23LC1024:
•
Length of bits: 524288'
Lots of detailed data