Deployment record
What 25 students reported, and what that does not prove.
Only M1, M2, and M3 have survey results. The classes were not randomized, and no class ran without the modules. The knowledge ratings are what students said about themselves. These counts are a deployment report. They are not a statistical result.
| Cohort | Term | Survey | Modules | Enrolled | Responses |
| C1 | Spring 2023 | Internal | M1, M2 | Not recorded | 13 |
| C2 | Spring 2024 | External evaluator | M1, M2, M3 | 13 | 8 (62%) |
| C3 | Fall 2024 | External evaluator | M1, M2, M3 | Not recorded | 4 |
C1, C2, and C3 in this table are the three classes. They are not the design rules C1 through C5.
The survey questions changed between Spring 2024 and Fall 2024, so those two classes should not be compared item by item. A response rate is known only for Spring 2024: 8 of 13 students. Of those eight, five were men, three were women, and six identified as Hispanic or Latino. All four Fall 2024 respondents were men, and one identified as Asian.
Self-reported knowledge, Fall 2024
Four-point scale. One is no knowledge and four is strong knowledge. These are self-reported ratings. There is no concept inventory behind them.
BeforeAfter
Binary reverse engineering
Read as 1.3 to 2.5 for binary reverse engineering, 1.8 to 2.5 for buffer overflow, and 1.0 to 2.0 for side channels. Interest in hardware security on a ten-point scale moved from 2.8 to 4.5 in this cohort. Spring 2024 reported 9.1 at the end and has no matching start value, so that figure is a terminal value, not a gain.
Spring 2023 usefulness ratings
Five-point scale. The midpoint is 3. Means sat below it. Students were not enthusiastic about the early handouts, which is part of why later versions added worked structure.
| Item | Mean | SD | N |
| M1 usefulness for learning binary reverse engineering | 2.77 | 1.25 | 13 |
| M1 contribution to assembly and instruction encoding | 2.69 | 1.20 | 13 |
| M1 adequacy of materials and lecture | 2.83 | 1.14 | 12 |
| M2 usefulness for learning buffer overflow | 2.73 | 1.21 | 11 |
| M2 usefulness for a prior understanding of memory layout | 2.36 | 1.23 | 11 |
| M2 adequacy of materials and lecture | 2.55 | 1.50 | 11 |
What students said the modules did
| Item | Spring 2024 | Fall 2024 |
| More hands-on work than in other courses | 87% yes | 100% yes |
| Modules helped with embedded-systems fundamentals | 87% yes | 25% yes |
| Changed interest in a graduate degree | 75% yes | 0% yes |
| Sought a hardware-security internship | 100% yes | Not asked |
| Wants a career in hardware security | 50% yes | Not asked |
| Wants more hardware-security modules | 50% yes | Not asked |
| Needs more time to finish the assignments | 75% yes | 25% |
| Workload against similar courses | 63% more | 75% about the same |
| Overall experience, ten-point scale | 7.6 | Not asked |
The split on fundamentals is the result to keep. Spring 2024 agreed that the modules helped them understand embedded systems. Fall 2024 mostly did not. Half of the Fall respondents reported that they did not attempt the assignments. The sample is small. The two cohorts should not be averaged into one success rate.
Barriers they named
Students chose from a fixed list: worked examples, lecture, study materials, time, and teaching-assistant support. Equipment was not on the list. Their answers are about teaching, not about missing instruments.
| Barrier | Spring 2023, M1, share of selections | Spring 2023, M2, share of selections | Fall 2024, share of respondents |
| Worked examples | 33.3% | 28.6% | 0% |
| Class lecture | 29.2% | 33.3% | 25% |
| Study materials | 16.7% | 19.0% | 25% |
| More time | 12.5% | 14.3% | 25% |
| Teaching-assistant support | 8.3% | 4.8% | 50% |
The Spring 2023 shares come from 24 selections on M1 and 21 on M2. Fall 2024 is four people. Spring 2024 was not asked this question.
Four lessons for the next offering
1Students wrote most clearly about the repair.
One Spring 2023 student called the buffer-overflow assignment among the most interesting work in the program, and described finding where execution had to resume so that later secrets would return in R7. M1 has no countermeasure. Students asked what binary reverse engineering is for.
2A desktop mitigation is not on the 8051.
The lecture names stack canaries and address-space layout randomization. Neither exists on the AT89C51. The repair has to be in the program, because the platform does not provide one.
3Worked examples mattered more than new equipment.
The 2024 overflow handout adds partial credit by how far the leak gets, and it adds step-by-step hints. Both match the barrier the surveys recorded: examples, lecture, and time, not instruments.
4The toolchain has to run on the machines students have.
Free responses include a student who could not run the vendor tools on a personal machine. The three deployed embedded modules share one 8-bit target, the AT89C51, and one vendor IDE, Keil. Whether the same labs carry to a 32-bit core is untested.