Hacker News Reader

Comment by Cieplak | Hacker News Reader

$100 billion is not unrealistic for typical defense budgets. Modern warfare is transitioning from lead to silicon. It takes 6 lines of VHDL to backdoor a CPU. Modern chips have billions of transistors. It would be irresponsible of any modern superpower not to ensure their nation's machines are secure. I am a US citizen, but if I were a Chinese national, I might be worried about things like Intel Active Management Technology. Maybe mutual defense is a good thing if it prevents escalation. I'm still outraged that members of my Congress proposed violent conflict with China as an appropriate reaction to a data breech, as if their own organization had never accessed Chinese systems without authorization. It's probably just theatrics of elected officials from states with large military presence, but how evil is it to propose killing people for sneaking into your computers which you failed to secure properly?

View on HN

> I'm still outraged that members of my Congress proposed violent conflict with China as an appropriate reaction to a data breech, as if their own organization had never accessed Chinese systems without authorization

This +1. We are supposed to be intelligent adults but this behaviour is more like schoolyard bullies that behave one way and then cry foul when another does the same.

Psychology, naming & shaming, PR, politicking, and propaganda are important parts of any conflict. Just look at the cold war. It's not hypocritical; it's part of the game.

While defense is surely an important by product of an indigenous semiconductor industry the simply truth is that China is now on the edge of collapsing back on it self.

China became an economic superpower because it had a huge labor force to offer cheaply and provided an environment in which international corporations could exploit (not necessarily with a negative confutation but rather the direct dictionary definition of the word) it easily.

They are coming to the point in which they need to import a very large amount of the components which is ending up in the electronics which are being made in China even basic stuff (e.g. high quality capacitors are still primary made in Japan, same goes for thin film resistors and other high precision electronics).

With electronics becoming nothing more than system on chip China needs to be able to compete locally if nothing than at least to offer cheaper local alternatives (A Chinese MediaTek for example), if they'll fail they risk at severely cutting the profitability and hence the income from Chinese manufacturing and risking quite a bit of it being actually shipped overseas especially when your production can be highly automated and scaled up without the need for a large and cheap labor force.

Another lucrative aspect that China has been offering is it's "design and manufacturing engineering" expertise you could go with a rough idea and a sketch to China and get 50 engineers all of who are experts in various fields of manufacturing from injection molding to electronic assembly and get your product out of the way.

To maintain this capability China needs to be able to design bespoke silicon just as easily as they can design a bespoke plastic shell which will be injection molded and then ultrasonically welded.

I was curious about how a CPU backdoor would work:

http://es.slideshare.net/ortegaalfredo/deep-submicronbackdoo...

This example is a memory inspector and poker. On a side note, I think using a LEON MCU would be smart for this kind of demonstration, the code is free and it's synthetisable in a lot of targets.

Another example[1] (actual paper[2]) of potential for cpu backdoors.

[1]: https://ece.umass.edu/news/paper-stealthy-trojans-attracting...

[2]: http://sharps.org/wp-content/uploads/BECKER-CHES.pdf

At minute 39:30, the lecturer demonstrates a snippet of HDL which will transition the CPU to ring 0 if a very specific instruction is passed to the processor core:

https://www.youtube.com/watch?v=fwyh3Vychj8

BTW, this kind of backdoor would be very visible in Bluespec. Bluespec is very concerned about resource use and you will have to make extra effort to bring (really unneeded) resources to be accessible during interesting stages of CPU pipeline.

Absolutely true, if you have the source code, which is trade secret. Correct me if I'm wrong, but Bluespec is like Haskell that compiles to Verilog? Seems pretty neat.

Any such thing as circuit decompilers? In my imagination, they look like as huge boxes with tunneling electron microscopes inside, where you input a chip and it outputs source code.

The closest I've seen to that are the guys that reverse engineered a 6502 at the transistor level and built a "perfect" emulator for it. http://www.visual6502.org/

I think even with modern computer vision techniques, it would be very very difficult to do this kind of thing automatically. If you were a big state with cheap labour, you could probably hire a load of people to do nothing but map traces all day.

@joshvm looks like we ran out of thread depth..

That's super cool. I wanted to show you this too if you hadn't seen it already:

http://zeptobars.ru/en/read/KR580VM80A-intel-i8080-verilog-r...

That is extremely neat. I wouldn't have thought you could build an 8080 in less than 1K lines of Verilog.

Network controllers with DMA would also be prime targets that could even be tailored for a specific victim's network. They can sit passively and monitor for interesting network traffic as well as sift through memory unimpeded. And when something needs exfiltrating, it can send out packets undetectable by the cpu.

Edit- oh, and dropping code into memory for the CPU to run is not unpossible either.

>Modern warfare is transitioning from lead to silicon

For some reason - this really made it sink in...

The cold war WRT electronics in the next few years is going to be very interesting. The dystopian cyberpunk landscape we read about in the 80s was utopian vs the reality of what we will see in the future.

> It takes 6 lines of VHDL to backdoor a CPU

Are they still using VHDL?

I wonder what hardware languages are being used by Intel and ARM.

VHDL is used primarily in Europe (and in some academic classes in the US), while Verilog is used by the vast majority of businesses in the US, and most of the top schools for VLSI. I know that the vast majority of chip designers I know (which are 90% US based) much prefer Verilog.

As for Intel and ARM, both are primarily Verilog shops, but I know ARM provides both Verilog and VHDL models for the licensed components.

Thanks. Looking at wikipedia, Verilog stems from 1984. I'm wondering why Intel and ARM don't develop their own hardware compilers. I mean, if Mozilla can decide to start a new language, then certainly can those big companies.

Like most languages, they evolve over time. The latest version of the IEEE standard version of Verilog is Verilog 2005, but there are also offshoots, such as SystemVerilog. While most things distributed for interoperability and targeted for general understanding are written in Verilog, most actual designers use SystemVerilog, which was eventually made into its own IEEE standard. SystemVerilog adds a lot of object oriented features, and is a pretty clean addition to Verilog.

As far as why Intel, ARM, and everyone else mostly sticks to these languages is similar reason to why C is still used... it is because it works, and works good enough. After 20 years of just having Verilog, enough people agreed that it needed a update, and SystemVerilog developed. Another big reason is the fact that Intel, ARM, etc. all rely heavily on scripts and design flows that are common between pretty much every semiconductor company... all of the EDA vendors have united around these standards, and in a world where interoperability is required and very little innovation happens on the backend tools, people stick to what works.

Intel used "iHDL" which was an in-house language from 1986 to 2005, and after 2005 switched mainly to Verilog.

You usually see VHDL in .eu and some defense contractors, and Verilog everywhere else. VHDL feels similar to Ada, and Verilog is more similar to C.

An interesting history from Intel's perspective is available here: http://webee.technion.ac.il/people/kolodny/ftp/IntelCADPaper...

A lot of companies still use VHDL to write IPs.

Systemverilog is the de facto standard in verification, but VHDL with its strong typing remains a favorite for writing synthesizable code.

ARMs customer facing code is 100% verilog and this reflects most of the industry.

They do provide VHDL models for at least some components I have seen, but Verilog is what they do seem to prefer)

Reply on news.ycombinator.com