1	Computer Hardware for the Post-von-Neumann era Ben Tabatowski-Bush Owner, EYCPH LLC Website: www.pcables.com
2	Order of presentation About the Author State of the Art for Turn-of-Millenium Computing: The von Neumann Architecture (VNA) Electric circuits in the cortex If von Neumann had lived longer Characteristics of the Post-von Neumann Architecture (PVNA) Coming in the next lecture in this series
3	About the Author Ben Tabatowski-Bush is an Engineer, a Vegetarian, and a pioneer of new millennium Hardware. He graduated with a BSEE from U-M in 1989 and with a MSEE from Michigan State in 1991. He worked for IBM in the early nineties in the Advanced Computer Architectures department working on multiprocessor machines. The highest performance computing machines on earth circa 2007 started in this location. Ben has been engineering alternative vehicle technologies in the Midwest for about 15 years (Zero emission vehicles, Hybrid vehicles.) He is an expert in the area of automotive controllers for these types of vehicles. Mr. Tabatowski-Bush is the owner of ECYPH LLC, a manufacturer of Hardware for mobile computing and supporter of the Post-von-Neumann Architecture (PVNA) movement.
4	State of the Art for Turn-of-Millenium Computing: The von Neumann Architecture (VNA) History of von Neumann architecture Features of von Neumann architecture Turing Machines and human intelligence What von Neumann knew of the brain
5	History of von Neumann architecture The modern computing era began in 1956 with a manuscript from John von Neumann titled “The Computer and the Brain”. This detailed the von Neumann Architecture. He died before it was completed. Very good reading. It summarizes the important concepts needed to build a Stored Instruction Machine. Any computer you have (circa 2007) utilizes the VNA or an immediate derivative.
6	Features present in von Neumann’s architecture A digital memory to hold instructions and data (which carries an attendant bottleneck between information and the hardware which processes it) An ALU to carry out math and Boolean logic A control unit to coordinate operations Means of moving data in and out (“I/O”)
7	Features present in von Neumann’s architecture (cont.) NOTE: The hardware technology of von Neumann’s day made memory relatively cheap and processing hardware quite expensive. Thus, an optimal cost architecture was heavy on memory and light on processing (ALU) elements. If you compare SRAM pricing to that of a central processor chip today, you’ll find that not much has changed. Hence the continuing popularity of the VNA with Finance and Accounting in our companies.
8	Turing Machines and human intelligence Alan Turing was a peer of von Neumann and was very influential in the birth of modern computing. The “Turing Test” and the “Turing Machine” are his best known contributions. A Universal Turing Machine (UTM) is a Turing Machine capable of emulating any other Turing Machine. The Von Neumann Architecture is an example of a Universal Turing Machine. Turing was able to prove that a Universal Turing Machine is incapable of solving certain mathematical problems such as proofs. You can look up Entscheidungsproblem from Hilbert.
9	Turing Machines and human intelligence (cont.) And yet, the human brain is somehow capable of such things as mathematical proofs, and so per Turing must not be representable by a Universal Turing Machine (UTM). But, any computing device which is deterministic and can be completely described by step-by-step instructions is representable by a UTM. This line of reasoning might lead one to the idea that the human brain (or human intelligence) is not deterministic, or might not be representable by a series of step-by-step instructions.
10	What von Neumann knew of the brain Von Neumann was very knowledgable about the human brain! (Check out his book on the topic.) His work on the brain was interrupted by his death and is the part of his body of work left incomplete. One can directly trace the roots of modern computer architecture to the ideas von Neumann had of the human brain. Von Neumann in his final work spoke of an internal “language” of the brain expressed in the periodic signals of individual neurons, bundled together in fibers. Such an arrangement allows for high-precision operations with low arithmetic and logical depth. He noted that this language is quite different from our conventional ideas of language.
11	Electric circuits in the cortex It turns out that human intelligence is largely related to the processing capabilities of neurons in the neocortex and their organization. Detailed study shows that these neurons behave as do analog circuits in many aspects. On a physics level, the neurons actually have many analog circuits in them (!) But, parts like synapses have many chemical mechanisms and are not electric circuits as such.
12	Electric circuits in the cortex (cont.) Linear and nonlinear summation/subtraction Multiply Logical (boolean) operations Integration Threshold comparison Delay element
13	Electric circuits in the cortex (cont.) A single neuron is capable of an astonishing assortment of very complex operations, far exceeding the capabilities of the simple ALU’s at the heart of the VNA. Further, much of the complexity comes from the action in the synapes, of which there can be ~6000 in a single neuron. These synapses are very small, and each one has computational capabilities still under heavy investigation. So, don’t believe the predictions of computers exceeding the ability of the brain in the next few years. These sorts of predictions way underestimate the MIPS rating of a single neuron.
14	If von Neumann had lived longer It’s ironic that von Neumann, who had such an interest in the function of the brain, developed his namesake architecture which is almost entirely unlike the brain. One is digital and massively centralized, the other is analog and massively decentralized. From Turing, we find that a VNA (a Universal Turing Machine) does not have hope for implementing intellegence as we conceive of it. I believe that if von Neumann had lived a bit longer, it was inevitable that he would’ve developed a PVNA as will be described next. Of course, if he had, this wouldn’t be described as a “Post von Neumann Architecture”
15	Comparisons of PVNA to VNA VNA emphasizes a small number of processing elements and a quite large memory with a small bottleneck access port. PVNA emphasizes an enormously large number of processing elements with learning and memory functions carried out by something called synaptic plasticity and the organization of connections between processing elements.
16	Comparisons of PVNA to VNA (cont.) VNA emphasizes high reliability, fidelity, and precision of memory storage through the use of digital technology and usage of Universal Turing Machine concepts. PVNA is much less precise in memory storage owing to the stochastic nature of syapses and the limited precision of the frequency of spiking neurons.
17	Comparisons of PVNA to VNA (cont.) VNA emphasizes the deterministic obedience of hardware to the programmer’s instructions PVNA is not so much focused on obeying a high level language as it is on observing data from the world and organizing itself in accordance to the natural symmetry and structure in the data, which then enables functions such as recognition and prediction (“intelligence”.)
18	Comparisons of PVNA to VNA (cont.) VNA would never be what we could call “intelligent.” – at least, that’s what Turing says. PVNA implements a large parallel structure of analog (or analog-like) elements which learns and operates on data from the world much as a cortex does. As it can carry out tasks previously performed only by a neocortex computing structure, it can reasonably be described as “intelligent.”
19	Coming in the next lecture in this series More details on implementing the low arithmatic/logical depth operations from the world of neurons in silicon-based analog and digital circuits How you can buy hardware from ECYPH LLC which enables you to carry out experiments in this exciting new field (or, “How to beat your competition to market!”)