This website contains materials from a past semester. Information, assignments, and announcements may no longer be relevant. Please refer to the current semester's site for up-to-date content.

Policies

⚠️ Warning
The following is just a summary.
Please refer to the Lecture 1 slides for the golden, most up-to-date source of truth.

Course Structure & Goals

This is a research-focused class designed to be hands-on with very little traditional lecture material. There is no textbook; students will learn through “struggling” and practical application.

  • Primary Goal: Prepare two designs (Physical AI accelerator for VLA applications and Single-Chip Micro-Mote V26) on a single 2.5mm x 2.5mm die (approx. ⅔ digital, ⅓ analog/mixed-signals) and send them for fabrication by the May 15th deadline.
  • The Chips:
    • NPU (Digital): Neural Processing ASIC for VLA-based Robotic Applications.
    • SCµM (Mixed-Signal): Single Chip Micro-Mote.
  • Format: The course begins with labs to assist with onboarding and tool learning. Afterwards, students work in assigned sub-teams to reach design goals using agile design practices.
  • Technology & Tools: We utilize modern FinFET technology (TSMC 16 Compact) and a mix of open-source and commercial tools, leveraging research flows from SLICE and BWRC.
  • Resources: If unfamiliar with the flows, visit ucb.bar or ucb.bar/cydocs.

Class Logistics

  • Lectures: There are only approximately 5 lectures. The remainder of the course consists of team updates and presentations.
  • Locations:
    • Combined Lectures (first lecture): Cory 540AB
    • Digital Lectures: Cory 540AB
    • Analog/Mixed-Signal (AMS) Lectures: Cory 293
    • Office Hours: Cory 111 (Schedule TBD)
  • Dropping the Class: If you decide to drop, please inform the staff immediately. We are here to support you and won’t ask for a reason if you don’t wish to share, but giving us a “heads up” is required to manage team assignments.
  • Participation: Do not fall behind. Staff may not have the capacity to help you catch up. If you become inactive or fail to provide consistent updates, you will be de-prioritized for tasks. Fewer tasks result in less data to evaluate your performance, which will significantly hurt your final grade.
  • Logistics & Material: Most course material, including recordings and slides, is located on the Spring 26 Drive.

DSP Accommodations

If you require accommodations, please submit your request through the Berkeley Accessible Information Management (AIM) portal as soon as possible. Staff will automatically receive your Letter of Accommodation (LOA) and reach out to you to discuss specific arrangements.

Note: Once submitted, the system will email your Faculty Notification Letter to the instructors. You do not need to email us personally unless you have specific concerns you wish to discuss privately.

Grading Policy

Grading is based on your delivery of promised work and contribution to the design and tooling. You will receive a midterm evaluation with a projected grade to serve as a concrete marker.

Grading Criteria:

  • Opinions of faculty and TAs
  • Contribution to design & tooling
  • Opinions of peers via self-evaluation forms
  • Lab grades and completion
  • Attendance

Core Philosophy: Did you promise and actually deliver? Care about the course and the work, not the grade. If you deliver on your promises, the grade will follow.

Research & Collaboration (New for this Iteration)

This iteration is fundamentally a research tapeout and differs significantly from previous years.

  • New Territory: We are using a new PDK (TSMC 16 Compact) with no prior working flows. The Digital design includes a fully new architecture built from scratch.
  • Staff as Collaborators: Instructors and TAs act as research collaborators. We will be learning the new PDK alongside you and might know as much about a specific problem as you do.
  • Expectations: Treat this as a collaboration. Do your research and attempt to solve problems yourself first. If something is confusing or feels wrong, share your solution and tell us immediately.
  • Feedback: As this curriculum is undergoing change, your feedback is vital to molding the future of EE194/290-7.

Confidentiality & Data Security

Strict adherence to confidentiality is required.

  • TSMC Materials: The TSMC 16C PDK and design materials are strictly confidential. Never disclose PDK material (PDFs, DRC decks, GDS, LIB/LEF files, SPICE sims) to anyone outside the course or you may face legal consequences.
  • Data Security: No files should ever leave (be downloaded off of) BWRC machines.
  • Communication:
    • All tapeout discussions must happen in the staff-created group chats on Google Chat.
    • No Direct Messages (DMs) between students regarding the project.
    • Preferred method: Start a thread in a public group chat and use that for 1:1 discussion.
    • If a private chat is absolutely necessary, you must include [email protected] in the group.

AI & LLM Policy

Large Language Models (LLMs) can leak confidential information to the public web.

  • Strict Prohibition: Do not put any PDK material, TSMC numbers, or TCL scripts into an LLM (e.g., ChatGPT, Claude, Cursor).
  • Physical Design/Integration Teams: Absolutely NO use of LLMs is permitted.
  • RTL/Software Teams: You might be able to use LLMs, but you must ask staff beforehand.
  • Approved Usage: If permission is granted, you must use Gemini on your Berkeley account (which has SLAs covering data privacy).
  • Urgency: Do not let deadline stress push you to violate this policy. If you are desperate, ping staff immediately—even late at night—rather than using an unapproved LLM. Flag the request as “URGENT.”

Academic Integrity

TL;DR: Treat the lab assignements like any normal project/homework from any other Berkeley EECS course. The rest of the course can be collaborative with code sharing permitted (aka: anything goes, as long as no NDA violations occur).

https://www-bsac.eecs.berkeley.edu/~pister/etc/Cheating.htm

https://eecs.berkeley.edu/resources/students/academic-misconduct/

Please note that obtaining, sharing, and posting solutions to any lab assignment is a violation of academic integrity. This includes uploading project code, official solutions, your own solutions, etc. to any site that is accessible by other people, such as a public GitHub repository.

All lab code and written responses should be original. We trust you all to submit your own work, but to protect the integrity of the course from anyone who doesn’t want to play by the rules, we will actively be checking for code plagiarism (both from current classmates and previous semesters). If you use a code snippet from a website like StackOverflow for a small task (for example, capitalizing a string), this is fine, but please cite your sources in your code with a comment. When in doubt, just cite your sources.

Using automatic code generators such as OpenAI Codex, Github Copilot, or ChatGPT is not allowed for the lab assignmenets unless explicitly permitted by the course staff.