Complex Chips Make Security More Difficult | Hacker News Reader

Complex Chips Make Security More Difficult

semiengineering.com

29 pointsPaulHoule17 comments

The hardest kind of attack to protect against is a security hole that is deliberately introduced by an employee who is actually a mole for a state actor. Frankly, I'm a little surprised that an attack like this has not yet been discovered because it would not be that hard to do (relative to the capabilities of state actors).

(This concern also applies to secure comms apps, BTW.)

The hole is that state actors can simply compel corporations to give up the data.

There are now trillion dollar corporations but the US government could shut them down tomorrow if they truly wanted.

Different states exist. That is, supply chain security involves a lot of possible state actors trying to subvert other state and private actors for a variety of reasons, from ransomware to intelligence to simple opportunism for chaos.

It has almost certainly already happened. If you program in C it is easy to introduce a buffer overflow and you can never prove it wasn’t an accident.

Chips aren't programmed in C. Well, maybe some of the firmware is, but the logic itself isn't.

"The practice of “zero trust” is particularly important here. Any device or software outside the organization cannot be trusted". Great approach but it is amazing how many breaches are still happening due to lack of basic prevention measures taken

> Any device or software outside the organization cannot be trusted

O_o

https://en.m.wikipedia.org/wiki/Zero_trust_security_model

"... sometimes known as perimeterless security, describes an approach to the design and implementation of IT systems. The main concept behind zero trust is “never trust, always verify,” which means that devices should not be trusted by default, even if they are connected to a managed corporate network such as the corporate LAN and even if they were previously verified."

Yes, lets make simple chips and transfer the security complexity into the compiler. ARMed to the teeth, they shall be.

Itanium vibes

Well Intel has a history even when chips were less "complex".

The only market penalty they ever got was for being behind in miniaturization, lagging behind TSMC and Samsung.

The only thing Intel management ever cared about for the last two decades was rewarding themselves with big payouts and compensation.

Kind of silly title. I think complexity makes anything more difficult.

Security will play a huge role in obsoleting software development as a job.

Monkeys in chairs papering over generic CPU design is pushing chip makers to consider silicon designed to workload spec; input parameter set, let it go.

Chips are now undergoing their great decoupling like software. It’ll take a while as manufacturing process pivots but rather than 8 generic cores we’ll eventually have SOCs per application. Software will be pushed to the UI layer alone for users, and whatever industry needs to boot strap manufacturing.

Frankly I’m looking forward to it; I can’t think of anything software companies have provided humanity that will stand the test of time, except making us all learn their new preferences.

So the whole reason we have general purpose CPUs is because in the 50's/60's they were hulking giant beasts both in size and power consumption - computing was only economical on these devices if shared.

Things shrunk and now everyone has a computer in their pocket - which in most cases is a window to the next generation of hulking giant beast - the "cloud".

So if I'm going to develop applications for the cloud instead of PCs or mainframes, it's going to need to be general purpose--because it's shared and rented by the minute/CPU cycle. Just like the mainframes of old. So some notion of general-purposeness will always be there.

Maybe hardware will support and fossilize around programming languages? One can argue that x86 is doing that and is essentially an "SOC" for C.

It seems to me that "general purpose" was a big driver of PC's and phones.

PCs are good for word processing, games, scientific computing, tracking satellites, etc.

The appeal of a smartphone is that it replaces a dumb phone, a walkman, TV set, video game console, watch, and many other things.

General purpose won as it was realized monolithic chips that did it all were infeasible due to manufacturing technology in the past.

Overtime the political story about market winners took over, and technology development became mired in MBA bean counting to extract wealth, as it became clear the public would happily consume what was fed them via PC screens like we did TV.

None of these textual objects or CPUs will mean anything in the future. Manufacturing technology will evolve to provide “end to end” computing gadgets, black boxes that need user interface code at most, to contextualizing outputs. We’re subsidizing a process of technical evolution at scale.

Describing this all through the words of contemporary political discourse is missing the point.

> pushing chip makers to consider silicon designed to workload spec

Well .. the big one of these is GPUs, which started as fixed-function pipelines and turned into massively parallel execution systems repurposed to cyrptocurrency.

Various people have tried to do niche things such as neural net coprocessors. General-purpose or GPU-like or DSP-like systems tend to win in the market ... because they're run on software, which is more flexible.

There's a reason ARM are now up to 180 billion cores shipped; the easiest way of making a custom IC is to wrap it around an ARM core with some firmware. The code is burned into ROM (which has security advantages and disadvantages!), but to me it still counts as "software".

At some point the iPhone will cross the triple digit of number of CPU cores onboard (counting across the whole board, not just the SOC and radio). It will be rather difficult for anyone other than apple to count when that is.

I’m going to rely on the insight of my friends in the chip biz working at Intel and Qualcomm back in the day; generic CPUs “won” as it became aware manufacturing anything different was literally impractical 10+ years ago.

The goal of custom to spec silicon never went away it just had to bide it’s time, then Intel got taken over by MBAs.

Choices were made based upon real data. That generic CPUs “won” due to consumer choice; anything winning due to consumer choice is just a handy political meme. We get what supply side can sell for the most profit.

Reply on news.ycombinator.com