A Brain/CPU Analogy?

I’ve been pondering Intel CPU specifications a lot over the last few weeks. Intel have made it mind-bogglingly difficult to work out just how a given processor will perform in a particular real-world application. With the variations in core architecture, bus width, bandwidth, parallelism and cache size, all interacting factors dependent on the nature of the processing task, it seems that benchmarking and modelling are nigh on meaningless. The only way you will ever know whether a particular CPU is going to benefit you is to try it. Take my recent experience – for most of what I do a Celeron Dual Core is indistinguishable from a Core 2 Duo at the same clock – who’d have guessed at that? I’m certainly feeling less compelled to rush and buy the latest and greatest CPU.

The more I looked at CPU specs, the more plainly I could see how my earlier meanderings about levels of memory tied in with the many levels of data accessibility in computer architecture.

Let us assume that the piece of the brain that we recognise as our consciousness is the ALU (Arithmetic Logic Unit). Here is all the logic and the high-speed working memory, the registers, used in actual computation. This is the fastest part of a CPU – and to achieve the speed it uses prime real-estate and most complex technologies (expensive!!!). These characteristics apply readily to the cortex – increased convolution of the brain surface, increased cortical area, is an indicator of greater intelligence across species. This is where immediate calculations and data manipulations are performed. Over this region we see the fastest activity, however the signals tend to be small – although fast on a per-operation basis, many tasks show little periodicity (look at pipelining, branch prediction, etc. to see these things are dealt with in CPU design)  – what we see as a gamma rhythm is, as expected from this model, is usually relatively weak and spacio-temporally scattered.

Prime analogy point 1 – Gamma corresponds to computation using registers (working memory) only.

When the data a CPU requires is not available in the registers, it next looks at cache memory. For the sake of simplicity we’ll go with single level cache for the analogy. Cache memory is special high-speed memory in conjunction with a cache controller that endeavours to keep frequently addressed data close at hand. The controller looks at the cache to see if the data is there, and if not, scores a “miss” and has to look at the much slower, but cheaper and more plentiful main memory RAM. If the cache controller finds the data in cache, then it is delivered to the CPU in a fraction of the time it would have taken from main memory. The communication between CPU and cache is slower than between CPU and register. A periodic communication at this level would correspond to the beta rhythm. Although lower frequency, these exchanges involve more energy, because it is common to read/write many registers worth of data in a single cycle. Beta EEG represents higher energy than gamma.

Prime analogy point 2 – Beta corresponds processes requiring access to cache memory access (short-term memory).

As already mentioned, if the data isn’t in cache, the memory controller will have to look out into main memory. Main memory relatively cheap and plentiful, but it’s also much slower than the CPU. In a relaxed state (nominal CPU load), the brain/computer can tick along quite happily with regular shuffles of memory between cache and main. A transfer from main to cache will usually include a lot of potentially useful data along with that specifically requested, thus the alpha rhythm is seen to contain more power than beta.

Prime analogy point 3 – Alpha corresponds to processes requiring access to main memory (medium-/long-term, often including complex procedures).

After main memory we start getting into various levels of mass storage. The first (and the last we’ll consider in this analogy) is the hard-drive with cache memory. Hard drive memory is cheap and abundant. Pretty much anything can be stashed away without worrying too much about the space it occupies or whether it will be easy to find later. Depending on the raw drive speed, the disk cache size, and the speed of the interface to the CPU, requests for data from mass storage can taking anything from milliseconds to minutes or longer. Just like the human mind/memory – some things we can pluck out of nowhere in a flash, other things will surface unexpectedly days after the question arose. Again, the energy levels involved are much greater at this level of storage than at those above – and EEG recordings show much more power at these lower frequencies.

Prime analogy point 4 – Theta and delta correspond to processes requiring access to mass storage (long-term memory), with some very deep memories (such as those that might be dredged up out of delta) corresponding almost to removable storage, requiring “manual prompting” to make the connections available.

The extent to which this analogy applies literally is uncertain. What I know of brain function does not preclude this model. As an AVS developer, however, such models are incredibly valuable. Imagery like this useful to keep in mind when designing a session to guide the brain in certain ways – it gives some sort of basis for determining how long to stay at a particular frequency, or when it might be beneficial to inject a burst of something else. AVS session design is at least as much an art as a science, but as the great masters of painting have shown with their attention to anatomy, some reasonable knowledge and understanding of the subject being portrayed is essential.

Always keep in mind when considering things like this that the signals we see in an EEG recording, the rhythms in which we are interested, do not exist. What we see is an aggregate of many, many single impulses, each too small to detect from the scalp. A brain rhythm is not suggestive that the brain is doing something with that period, but that it is doing many, many more-or-less synchronised somethings so that what you see is somewhat akin to a Mexican wave.

Useful and not blatantly wrong is often as good as it gets at the bleeding edge.

Cheers,
Craig

Advertisements
Post a comment or leave a trackback: Trackback URL.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: