Oooh! Very cool stuff. As one whose academic background was in chemistry, I really like that.

I've heard interesting things about the RISC-V architecture which have made me want to look more into it. Logic design is something that I feel isn't nearly touched upon as much as it should in other technical disciplines. My undergraduate engineering coding experience was complete shit for lack of a better term.

Yeah. I've been slowly working towards going through Learning the Art of Electronics but, too much has come up this year. RISC-V, coupled with artificial hardware shortages (scalpers), is really what got me interested in digital circuits, as I've been a FOSS proponent for years and have been gaining both annoyance at lack of innovation and effort to rectify supply shortages, as well as less and less comfortable with centralized control of hardware implementation (repairability, privacy, modifiability, and general ability to figure out how a thing works).

RISC-V, to me, seems to be solving a lot of that by encouraging more open approaches to ISA development. It isn't quite GPL level but, some of the most performant and general-purpose oriented implementations have been open-sourced by their implementers (ex Berkeley BOOM series of cores and even Alibaba's THead cores, which are likely to be present in servers by the end of the decade).

The biggest challenge, to me, is that there are not currently any great ways to make the ISA implementations democratized, accessible, and performant in modern use cases. The performance side is directly related to the two options for implementation: FPGA and custom silicon.

FPGAs are limited both in number of logic cells and physic due to the size and nature of their logic gates (bigger and further apart means slower and less efficient). For example, an implementation of a modified Berkeley SonicBOOM (BOOMv3) that is hardened against Spectre attacks takes about 115k LUTs per 100MHz core (my Xilinx 7-series FPGA, which is 4 generations old, only has an 85k LUT capacity, a cutting-edge FPGA chip costs thousands of dollars just for a bare chip).

Custom silicon is just insanely expensive and out of reach for most. The lithography equipment necessary for currently node processes is pretty much only available to multinational corporations. I'm going done alternatives that are "good enough" for modern workloads become more available.

I think that having a masterclass to get everyone on board with using Vim/Neovim/Emacs (Something at least that has much more versatility as an editor) as well.

Absolutely. That sounds like a worthwhile undertaking.

Embedded systems have had particular interest for me too. I took a control theory class in my senior year where we had to analyze reactors, distillation columns and disturbances in steady states: (i.e. fluctuations in P,T,x_{i} (liquid fraction compositions), and many more). I'm assuming the embedded systems that actually implement this could go into the curriculum too.

Indeed. I'm mainly self-taught there, with focus on hobbyist MCUs but the entry requirements both financial and knowledge have gone down significantly. An RP Pico W, a dual-core 32-bit ARM MCU with onboard WiFi and BLE can be had for $6. It can also be programmed on bare C, Arduino C, at least two implementations of Python, Rust, and a number of others, including drag-and-drop languages like MakeCode.

Definitely interested to see where this goes and contribute.

The definition of "HPC" is actually a constant point of contention in my job. Most scientists and indeed most management people I work with use the term "HPC" to describe any multi-tenant science computing infrastructure that uses a Job Scheduler, regardless of how HP the C actually is. Most of the time I actually avoid the term, since it's got this ambiguity.

The system I work on takes up three full racks in the datacentre, but in the grand scheme of things it's not that big of an HPC, only 20 compute nodes of around 64 cores and 1-3 TB of RAM.

Most of the work my clients are doing is in the realm of bioinformatics for research purposes, so things like genomics, metagenomics, etc. Not so much something like protein folding. Mostly it's lots of big complicated data science pipelines where the inputs are things like FASTA files and reference data. Lots of machine learning and statistical models trying to make connections between plants, animals, bacteria, fungi, viruses, etc. Notably, the science my organisation works on has very little to do with Human biology specifically.