Atoms

Transistor, Application, Fabrication (2022 Update), CPU - OS (2023 Update) + ChatGPT Sampling,
(2025 Update on GPU)

1. Transistor (an introduction to its principle of operation - turning the electronic parts on and off at certain voltage) -
   
   Solid state transistor is ubiquitous in modern life since the 1950's when its usefulness was discovered at Bell Labs in 1947. As shown in Figure 13-03c(a), this device consists of all the three types of solid in the Band Theory. The semiconductor is usually a piece of doped silicon, which allows some free electrons (N-type) or holes (P-type) to move around. The transistor is off admitting

   	at most a weak current before a certain threshold gate voltage VG (usually a few volts) is attained at the base. Once the gate voltage is over the threshold voltage VTH, the transistor is turned on (acting as a switch) and the current increases exponentially (acting as an amplifier, see Figure 13-03c(b)). Figure 13-03c(c) shows the effect of VG on the valence and conduction bands. The
   Figure 13-03c Transistor [view large image]	bendings of those are ultimately responsible for turning on the current flow in the transistor when VG - VTH > 0.

   
   

   	This type is called Field-Effect Transistor (FET) and uses mostly in integrated circuits (IC). There are other types of transistor such as bipolar junction transistor (BJT). It is especially useful as amplifier in analog circuits (see Figures 13-03c2, 13-03e, and "Difference between BJT and FET").
   Figure 13-03c2 BJT/FET

   		Figure 13-03d shows a very primitive electronic circuit (characterized by non-linear relationship between current and voltage) for testing a BC547 transistor with many components including : Diode - A semiconductor diode, is a P type semiconductor connected to a N type one with two electrical terminals. It allows current to pass through in only one direct (a rectifier). LED (light emitting diode) is a diode emitting light at the junction. Resistors - They are used for protecting the transistor and diode from damage by current overload. Pot - The potentiometer is an instrument to measure the voltage of a seat without drawing any current from the seat.
   Figure 13-03d Electronic Circuit	Figure 13-03e Logic Gate [view large image]

   Beside acting as switches and signal amplifiers, transistors are also involved in power regulators and equipment controllers. They are the building blocks of integrated circuits and most modern electronics. Microprocessors often include more than a billion of them in each chip. They are in fact used in almost everything, from stoves to computers and pacemakers to aircraft. Figure 13-03e illustrates just one of the applications as a logic gate (the component in the CPU - Central Processing Unit). This one performs the function of the AND gate, which produces an output signal only when both inputs are turned on.
Other electronic components include :

• Capacitor - Filtering and frequency modulation, etc.;
• Triode -Amplifying, switching, etc.;
• Crystal Oscillator - Providing square wave or sine wave;
• Transformer - Changing the voltage;
• Connectors - Connecting two active components to carry the current or signal, etc.
The difficulty of fabricating Integrated Chip (IC) is due to the extremely tiny size of the primary component - the transistor.


2. Application - The digital age flourishes partly due to the representation of image and sound by series of binary numbers 0 and 1
   

   	(see Figure 13-03f), and also thank to the limitation of human sight and hearing. The ability to distinguish two very close points is about 0.01 cm, and the perception of image persists for about 1/16th of a second (Due to this, when many still images are shown in a
   Figure 13-03f Digital Age [view large image]	sequence; they give the illusion of moving images. For example, when separate images are moved in front of the eyes at the speed of at least 16 frames per second, we experience the images being real and live). The following example is mostly about visual image.

   
   
   • Size Scale of the Integrated Chip (IC) - The smallest addressable unit (by software) is a pixel with linear dimension ~ 3x10^-2 cm including spacing (Figure 13-03g). Each pixel is made of 3 sub-pixels, for Red, Blue and Green colors in 1 byte length (= 8 bits) each. The binary value in each byte corresponds to the intensity of the color, for example, "11111111" represents full load while "00000000" indicates its absence. In this case, there would be 2⁸ = 256 different shades of colors. As it will be explained later, ultimately the shade is determined by the gate voltage V_G to a transistor (see Figure 13-03c).
     

     		Figure 13-03h zooms into the IC to show its innards in size scale ~ 100 times smaller on each successive frame : (a) IC in package with linear dimension ~ 5 cm. (b) Interconnecting tracks under the IC with size ~ 10-2 cm. (c) Further enlargement of the tracks ~ 10-4 cm. (d) An image of the transistor ~ 10-6 cm.
     Figure 13-03g Pixels [view large image]	Figure 13-03h Size Scale of IC [view large image]

   [2024 Update] :
   
   Now that the chip is getting smaller to the nano-meter scale as shown above; it reaches some physical barriers which requires ingenuity to overcome. ChatGPT has provided a summary below (in Italic text) :
   
   1. Limits of Chip Miniaturization
   Smaller chips (below the nanometer scale) face a range of challenges:
   " Quantum Tunneling: As transistors shrink, the insulating barriers between them become so thin that electrons can tunnel through, leading to leakage currents and inefficiency.
   " Heat Dissipation: Higher clock speeds and greater transistor density mean more heat per unit area, which is difficult to dissipate effectively.
   " Signal Integrity: Smaller transistors and interconnects mean higher susceptibility to noise and signal degradation.
   ________________________________________
   2. Approaching Physical Limits
   Modern chip designs (like those at 2 nm or smaller) approach the atomic scale, where classical semiconductor physics transitions to quantum effects. This doesn't mean it's impossible to make smaller chips, but:
   " New materials (e.g., graphene, 2D materials) and structures (e.g., FinFET, Gate-All-Around transistors) are required.
   " Technologies like quantum computing, which uses qubits instead of classical transistors, offer alternatives to traditional chip design.
   ________________________________________
   3. Potential Solutions and Alternatives
   " Low-power designs: Innovations like better thermal management, reduced operating voltages, and efficient architectures.
   " Parallel computing: Instead of increasing clock speed, chips now rely on multicore designs and parallel processing.
   " 3D stacking: Chips can expand vertically, offering higher performance without further miniaturization.
   " Post-Moore's Law technologies: Approaches like photonic chips and neuromorphic computing could bypass the limits of traditional silicon.
   See "Single Photo (2024)"
   ________________________________________
   
   In summary, while chips are indeed approaching physical and practical limits, continued innovation in materials science, chip architecture, and alternative computing paradigms will drive progress beyond the current barriers. The uncertainty principle isn't a direct limiting factor for heat or size, but quantum effects do become more prominent as chip features shrink.
   
   The most advanced semiconductor chips are being developed by a few leading manufacturers:
   
   1. TSMC (Taiwan Semiconductor Manufacturing Company): TSMC leads the industry, producing over half of the world's semiconductors. It is the only company, alongside Samsung, capable of manufacturing chips using the most advanced 3nm process node, and it is progressing towards 2nm and even 1.6nm-class nodes by 2026. These advancements leverage cutting-edge technologies like nanosheet transistors and backside power delivery for improved energy efficiency and performance
   2. Samsung: Based in South Korea, Samsung is TSMC's closest competitor. It also produces 3nm chips and is pushing advancements to compete in high-performance computing and mobile applications
   3. Intel: While not as dominant in foundry services, Intel is investing heavily in its advanced node development. Its roadmap aims to regain technological leadership in the next few years by focusing on innovative manufacturing and design technologies
   4. GlobalFoundries (A foundry is a factory that produces metal castings) and UMC: While they focus on less advanced nodes compared to TSMC and Samsung, these companies are crucial players for manufacturing in sectors that do not require cutting-edge chips.
   
   These companies are at the forefront of technological innovation, pushing the limits of what is possible with semiconductor fabrication. TSMC and Samsung's dominance is driven by their ability to invest heavily in research and development and maintain advanced production capabilities.
   
   [End of 2024 Update] :
   
   
   • Image Scanning - The pre-digital era used to capture image on film which has three layers of emulsion, each layer is sensitive to a different color, it records on each tiny spot of the film to reproduce the same color as the image projected onto it, same as the lens perceives and the human sees. The following uses the modern version of the camera to illustrate how it is done in digital age. Essentially, it converts the analog image to binary numbers. Instead of using the film to keep the image, it stores the digitalized image on a digital file. The image is retrieved by more electronics to display it again digitally in such a way that we consider it to be as good as the original. The various panels (from "a" to "d") in Figure 13-03i explain in some details about such modern process (see "A few scanning tips").
     
     (a) This is a high level view of the process. It uses a mosaic filter to capture the colors of the image. The basic element is the pixel (or photosite in camcorder's lingo) as described earlier or a variation (called Bayer pattern with 2 green sub-pixels). This pattern is transformed to digital file by the analogue/digital converter (ADC).
     

     (b) This one shows the 2 different designs of the image scanning systems - CCD (Charge Coupled Device) and CMOS (Complementary Metal Oxide Semiconductor) image sensors. While CCD goes through the photon-to-electron (involving the capacitor C) and electron-to-voltage (as the amplifier A, both C and A are FET) steps within the image sensor before running the ADC process in an circuit board, the CMOS has these steps completed in the pixel and the ADC is executed within the image sensor (see further information in "Background Information on CCD and CMOS Technology"). (c) Further elaboration on CCD and CMOS, especially on the charge transfer step in CCD in which the electrons are moved from one register (pixel) to another serially.

     Figure 13-03i Image Scanning [view large image]

     (d) This is a pictorial displays about photon-to-electron in the FET, and electron transfer by shifting the applying voltage (to the FET) in CCD.

     For video scanning, the photosites reset to start the exposure to the next frame after the conversion is complete.
   
   
   • Storage - DVD (Digital Video Disc) appeared in the market near the end of last century allowing up to 17 GB of storage and gradually replacing the analogue cassette tapes invented back in the 1960's. The capacity of DVD has improved over the years up to 128 GB in Blu-ray format (Figure 13-03j). The decline of DVD technology is not so much about capacity as

     Blu-ray disc has enough space to keep so many copies of movie; it is the advant of broadband internet streaming (also see "List of streaming media services"), which skips the requirement of storage as opposite to downloading. Anyway, the Blu-ray and its like are still somewhat useful today (as of 2022) for keeping a copy of whatever. However, the USB (thumb drive) is a preferable alternate. Since the mid 2010's, the laptop computers

     Figure 13-03j Digital Storages [view large image]

     and audio port in automobiles have the DVD slots replaced by USB terminals. Followings are brief description of the digital elements in various storage media.

     (a) Compact Discs has gone through many incarnations from CD to DVD, HD-DVD, and now Blu-ray. As shown in
     Figure 13-03k,a, the size of encoding element (in a piece of aluminum) decreases, and hence both the data density and capacity increase (from 0.7 GB) in each succession.
     CDs are made from an original "master" disc. The master is "burned" with a laser beam that etches bumps (called pits) into the aluminum surface. A burned bump represents the number zero. The reflective spot is the unburned area (called land) representing the binary number 1. Thus, the laser can engrave all the information sampled from the original track of music or image in series of 0's and 1's on that thin piece of aluminum. In the read-back process (see Figure 13-03k,b, the laser (moving in and out from the center) flashes up onto the shiny (under) side of the CD, bouncing off the pattern of pits (bumps) and lands (flat areas) on the disc. The lands reflect the laser light straight back, while the pits scatter the light.

     	Every time the light reflects back, the attached photocell detects it, realizes it has seen a land, and sends a burst of electric current to an electronic circuit that generates the number one; otherwise, the electronic circuit generates the number zero. The series of data on the disc is decoded by an electronic circuit (called digital to analog converter or
     Figure 13-03k Compact Discs [view large image]	DAC) in the CD player and converts them back into a changing pattern of electric currents. A loudspeaker transforms the electric currents into sounds (by changing their electrical energy into sound energy).

     
     

     (b) As shown in Figure 13-03l, the Hard Disk Drive (HDD) Read/Write (R/W) mechanism is similar to those employed for Compact Disc (CD), except that the binary numbers are represented by magnetized spots on the disk. A typical HDD consists of a spindle that holds the platter. The platter is made from a non-magnetic material, usually aluminum alloy, glass, or ceramic. It is coated with a shallow layer of magnetic material typically 10-20 nm in depth, on which the digital data are magnetized with opposite magnetic field direction (related to the pre-set field) as 1 and otherwise as 0 (see Figure 13-03m,b).

     [view large image] Figure 13-03l HDD R/W

     (c) The USB flash drive (available since 2010) is a data storage device using flash memory (see "How USB Flash Drive Work?") to store data with an integrated USB interface (see Figure 13-03m). It is typically re-movable, re-writable and much smaller than an optical disc. The flash memory is similar to the Field Effect Transistor (FET, see Figure 13-03c,a), except that the insulator under the "Gate (now called Control Gate)" is replaced by the "Floating Gate" which stores electrons by applying voltage to the "Control Gate' creating the binary number 1 ("0" is represented by no charge). The charges can be removed by applying high voltage to the source (see Figure 13-03m and Figure 13-03n,c).

     Figure 13-03m USB Flash Drive [view large image]

     	(d) Solid-State Drive (SSD) uses flash memory similar to the USB flash drive. As such, it has no moving parts unlike the HDD and its storage capacity can surpass the HDD too (Figure 13-03j). See "What Is a Solid-State Drive (SSD)?" for further detail.
     Figure 13-03n Storage Element [view large image]

   • Reception of signal involves transmission from the sender and a device for its retrieval (see Figure 13-03f,c). Lot of progresses have been made for both aspects in the 21st century. here's a brief introduction mostly about how they work.
     
     (a) The concept of broadband internet depends heavily on frequency modulation for speedy delivery of data (aka signal, information, ...). Figure 13-03o shows a very simple example of frequency modulation with a single channel at low frequency. The whole idea is captured by the formula :
     
     where A_m = amplitude of the message wave (in blue color, under subscript m), f_m = message frequency, _m = 2f_m, f_c = carrier frequency, _c = 2f_c, modulation index = f/f_m, f the frequency deviation (from f_m), and bandwidth = 2(f+f_m).
     Note 1 : The abscissa in Figure 13-03o, diagram (a) denotes time t in unit of second and represented by Eq.(1), diagram (b) is in unit of frequency f (Hz = # of oscillations per sec) and diagram (c) is in unit of number of side-bands.
     Note 2 : The modulation of the carrier frequency is via the last term in Eq.(1). It appers as a time varying (in sine wave form) phase change to modify the carrier frequency f_c which is usually much higher than f_m.
     Note 3 : It is the side-bands which deliver the messages. The carrier wave is suppressed in "Reduced-carrier transmission".
     Note 4 : The side-band amplitude depends on f as shown in the table below left :
     Note 5 : Binary data of 0 / 1 are encoded as slightly higher and lower frequencies in the side-bands, see graph below right :                           
     
     By 2022, viewing of video is no longer a very slow process by downloading with a modem. Now the broadcast comes on

     		the screen instantly without delay, thank to the 19th century technique called "multiplexing" (more precisely "frequency multiplexing in this case) which is known now as "broadband internet". As shown in Figures 13-03p and 13-03q, the data are separated into several packages (called channels) near the carrier frequency. The information bearing wave would have certain bandwidth after modulation (see Figure 13-03o). These "data waves" would travel different routes in parallel
     Figure 13-03o Frequency Modulation	Figure 13-03p Broadband Internet [view large image]	making the delivering time much shorter. They are finally reassembled at the receiving end to be viewed.

     In practice, There are 2 carrier frequencies of 2.4 and 5 GHz in use by the public. The 2.4 GHz has 14 channels for bandwidth of either 20, 40, or 80 MHz (Figure 13-03r). Use of the 5 GHz with bandwidth up to 160 MHz may encounter problem with channel interference as the separation between them is reduced by about 1/2 (as compared to the 2.4 GHz), although data transfer rate would improved as it depends on bandwidth, i.e., broader bandwidth (hence "broadband") packs more data.

     Figure 13-03q [view large image] Broadband Transmission

     	Overlapping affects 2.4 GHz as well; therefore, only channels 1, 6, 11 are typically used in the US. In the rest of the world, the four channels 1, 5, 9, 13 are typically recommended (see "Wireless Radio Channels", and "2.4 GHz Channel Planning"). Figure 13-03r explains the concepts of carrier frequency, distribution of channels (and the overlappings), and bandwidth.
     Figure 13-03r Frequency Channels	The insert at upper right corner is an one channel packet with signal near the carrier frequency fc , the Side-Band (SB) within the Band-Width (BW) can also deliver data . Each carrier frequency is modulated as shown in Figure 13-03o.

     
     
     Followings are brief descriptions (with references) on some special terminologies in the cyberspace for the novices :
     
     Speed of the broadband internet depends on bandwidth (to pack data) as well as number of channels (for dividing data into packets in transmission and reassembling them at the receiving end). The example in Figure 13-03r has bandwidth of 22 MHz, which would be advertised in the trade as 20 Mbps Max with bps standing for "bits per second" or bits/sec.
     
     Common mediums of transmission include fiber (see "Optic Fiber"), wireless, satellite, digital subscriber line (DSL), cable modem, and broadband over power lines (BPL).
     
     Mobile broadband receives the data package directly from radio transmission (actually via the various plans offered by Internet Providers, see "Cell Phone Plans"). Similar service is also available via fiber cable connection (Figure 13-03s,a). These services are provided only in big cities where cost-profit margin is favorable for installing the fiber network.
     See the latest 5G networks for Mobile Broadband.
     
     WiFi Hotspot is free WiFi service (with or without a password, but with no security) offered to the public by stores, offices, ... via a router.
     

     	Digital Subscriber Line (DSL) also provides fast data transmission over copper telephone lines (instead of the even faster fiber cable). The "A" in ADSL refers to "Asymmetric" between more downloads than uploads (Figure 13-03s,b).
     Figure 13-03s Internet Service	Dialup internet use 20th century technology with modems to convert data from digital to analog and back to digital at the sending and receiving ends of a telephone line respectively. It is now used mostly in rural areas.

     
     Media Access Control address (MAC address) is a hardware identifier that uniquely identifies each device on a network. Primarily, the manufacturer assigns it. A MAC address can also be referred to as a burned-in address, Ethernet hardware address, hardware address, or physical address.
     
     Meanwhile, an internet service provider (ISP) or network administrator assigns an Internet Protocol address (IP address) to the client. Associated with the TCP/IP protocol, an IP address identifies a device connected to a network. It can be linked to an easy-to-remember domain name (name of the website as known in the street, see Figure 13-03t).
     
     

     	IP address protections : Use a proxy server. Use a virtual private network (VPN). Use a reliable antivirus software and keep it up to date. Stay alert to phishing emails and malicious content. Change privacy settings. Create unique passwords.
     Figure 13-03t IP Address - Domain Name

     
     (b) Finally, the signals reach the destination. Here's a brief description to explain how they become the image on the screen :
     
     

     	For those of us old enough to remember the CRT monitor in the 20th century. It was a chunky and very heavy object occupying 1/3 of the desk space (look like the image on the left). It was replaced by the plasma monitor in the early 2000's and followed subsequently by the LCD, and now OLED.

     The LCD display relies on a special property of certain Liquid Crytal (LC) to alternate the polarization of light. It is in turns controlled by the on and off of the ultra-tiny transistors. Followings describe the sequence which determine the state of a pixel in the displaying panel (see Figure 13-03u).
     
     
     • The light source is un-polarized from the back of the panel (hence called Backlight).

     The 1st polarizer admits only vertically polarized light to go through. In the absence of voltage V, the liquid crystal would rotate the light from vertical to parallel polarization permitting it to go through the 2nd parallel polarizer (see Figure 13-03a,c). For V 0, the liquid crystals are aligned with the field direction, and let the passing light to go through. It is eventually blocked by the 2nd parallel polarizer, the screen becomes dark (Figure 13-03b). The on and off of the voltage is controlled by the TFT (Thin Film Transistor)

     Figure 13-03u LCD Display [view large image]

     which is in turn controlled by the incoming signal. Figure 13-03d shows a small section of the LCD rectangular array. The RGB filters enables the display of vivid color.

     LED-backlit LCDs are often marketed as LED TVs. The truly LED TV uses OLED technology where the "O" stands for some Organic molecules which can emit certain color of light by applying a voltage (again controlled by the ultra-tiny transistors). A little bit more details :
     
     

     Some organic molecules have partially free electrons (called Delocalized electron) roaming between them. This property makes the system behaving like a semiconductor with some electrons staying in the conduction band. Applying an electric potential on this system returns some electrons to the partially filled valence band and emitting a photon in the process. The energy of this photon (i.e., the frequency or color of the emitting light) depends on the energy gap between the 2 bands (see Figure 13-03v,a for a single sub-pixel, diagram b displays a LED pixel). The bi-layer design allows different conducting property in 2 zones to provide a more gradual electronic profile (Figure 13-03v,c).

     Figure 13-03v OLED Display [view large image]

     See "What's So Good About OLED" and the next generation PeLED. Figure 13-03v,d is reminiscence of the first flat screen technology which uses different kinds of plasma to excite phosphors to emit light.

     
     
3. Fabrication - Manufacturing a microchip requires several steps summarized below (see corresponding images in Figure 13-03w) :
   
   

   	Figure 13-03w	IC Fabrication [view large image]

   
   

   Pre-treatment and Wafer Forming - The most common method for silicon wafer preparation involves preparing a single crystal seed and dipping it into molten silicon. Once in the liquid, the seed is pulled from the molten silicon slowly, while the rod rotates (Figure 13-03x). Then, the finished wafer is polished and has a crystallographically flawless surface. The circumference of the monocrystalline ingot is ground down to a uniform diameter. Then the sliced wafers are polished by alumina abrasive in a lapping machine to the desired thickness, while improving the surface parallelism.

   Figure 13-03x IC Wafer Preparation

   Cleaning - The purpose of silicon wafer cleaning is to maintain a clean and dry surface. The adhesion between substrate and resist (a resistant substance) is of major importance for the safe processing of photoresists (light-sensitive coating).

   
   
   • Oxidation - This process creates an SiO₂ layer, which serves as an insulating layer that blocks leakage current between circuits. The oxide layer also protects the silicon wafer during the subsequent ion implantation and etching processes.
   
   

   Deposition - This is the process to deposit thin films of conducting, isolating or semi-conducting materials (depending on the type of the structure being made) on the wafer to enable printing of the first layer on it. One of the method is Chemical Vapour Deposition (CVD) which deposits material in hotter gas phase onto a cooler surface (see "Explained: chemical vapor deposition", "Virtual Chemical Vapor Deposition Lab", and Figure 13-03y).

   Figure 13-03y Deposition Process

   Chemical mechanical polishing/planarization (CMP) - This is a process that removes materials by a combination of chemical and mechanical (or abrasive) actions to achieve highly smooth and planar material surfaces.

   
   
   • Ion Implant - Ion implantation is a surface treatment process in which ions of nitrogen or carbon are accelerated and made to penetrate the thin film endowing it with wear resistance.
   
   

   	Photoresist Coating - Photoresist (also known simply as a resist) is a light-sensitive material used in photolithography and photoengraving to form a patterned coating on a surface. Lithography - Lithography is the process of transferring patterns of geometric shapes in a mask to a thin layer of radiation-sensitive material (called photoresist) covering the surface of a semiconductor wafer.
   Figure 13-03z Process of Engraving	Etch - In semiconductor device fabrication, etching refers to any technology that will selectively remove material from a thin film on a substrate and by this removal create a pattern of that material on the substrate.

   The 3 steps above engrave a design on the surface of the wafer (Figure 13-03z).
   
   
   • Wafer Probe - A wafer prober is a system used for electrical testing of wafers in the semiconductor development and manufacturing process.
   
   

   Packing - This final fabrication process gets the chips out of the wafer, it is sliced and diced with a diamond saw into individual chips. The chips are cut from a 300-mm wafer, the size most often used in semiconductor manufacturing, these so-called 'dies' differ in size for various chips. Some wafers can contain thousands of chips, while others contain just a few dozen. Figure 13-03z2 illustrates why Integrated Chip (IC) becomes a hot topic in this digital age (see "Three months, 700 steps: Why it takes so long to produce a computer chip" by Washington Post, 2021; and "Causes of global chip shortage").

   Figure 13-03z2 From Wafer to Consumer

   Here's a YouTube video on "How ASML, TSMC And Intel Dominate The Chip Market". ASML - Advanced Semiconductor Materials Lithography. TSMC - Taiwan Semiconductor Manufacturing Company = 臺積電

   
   
4. Central Processing Unit (CPU) -
   
   

   Basic Elements - Conventional computers process information by breaking it up into its component bits and operating of those bits a few at a time. These computers consist primarily of electronic circuits including bits, wires, and gates. Bits can be implemented by ferrite cores (in memory), magnetic spots (in hard-disk), or the on and off of the voltages. These bits can be sent along wires to the logic gates for processing. It has been shown that any desired logical expression, including complex mathematical calculations, can be built up out of the OR, AND, and NOT gates (see Figure 13-03z3, and an example using NAND).

   Figure 13-03z3 Logic Gates [view large image]

   The same figure also shows the corresponding electronic circuit to implement the logic gate (just one example for the AND gate). Logic gates are primarily implemented using diodes or transistors acting as electronic switches. See "List of Logic Gates".

   
   
   • Hierarchy of Memory - As shown in Figure 13-03z3, all the logic gates requires binary input(s) to operate.

     	The inputs are kept in storages initially far away from the CPU when they are not needed. Those that are required for operation has to be housed in a place called register which can response as quick as the clock cycle (see Figure 13-03z9).
     Figure 13-03z4 Hierarchy of Memory [view large image]	Here's a summary of the various kinds of storage according to capacity and speed (also related to cost - more expansive for the speedier ones, see Figure 13-03z4).

     • Register - Registers are data holding places as integrated parts of the CPU (see Figure 13-03z4). It is a special kind of

       memory providing the quickest input/output of data (to the CPU). Its structure is very simple to hold 8, 16, 32 or 64 bits of binary numbers. It is usually implemented as "Flip-flop and then Latch" by 4 BJTs. Figure 13-03z5 shows a single element holding 1 bit of data which is set to 0 or 1 by a pulse of electric current and then holding it at that state until the next clock cycle for executing another instruction (see Figure 13-03z5).

       Figure 13-03z5 [view large image] Processor Register

       There are many types of register, each one has a special name in Assembly language (see for example, "Assembly - Registers").

     • Cache Memory - It stores a copy of the data/programs which are used very often. The high-speed memory retrieval would improve the performance of the CPU. It has 3 different levels as shown in Figure 13-03z4.

       Cache memory is implemented by the Static Random Access Memory (SRAM), which follows the same architecture for register but using 4 FETs instead (see the "difference", and Figure 13-03z6). It helps the processing because of no need for searching the information everywhere (if the information is in there already). Caching can also be applied and leveraged throughout various layers of technology including Networking layers, Content Delivery Networks (CDN), DNS, web applications, and Databases.

       Figure 13-03z6 Cache [view large image]

       Some computer also add buffer (a temporary storage) to queue up data for synchronization with the faster cpu processing.

       
       
     • Random Access Memory (SRAM/DRAM) - RAM is computer's short-term memory. In the most common form of computer memory, dynamic random access memory (DRAM), a transistor and a capacitor are paired to create a memory cell, which represents a single bit of data. The capacitor holds the bit of information - 0 or 1. The transistor acts as a switch that lets the control circuitry on the memory chip read the capacitor or change its state. However it is volatile, meaning that the contents would disappear once the power is turned off. There are two main kinds -

       	the Static DAM (SDAM) used for cpu cache needed constant power supply (hence Static) and Dynamic DAM (DRAM) used for fast primary storage needed intermittent power refreshment (hence Dynamic). See "Difference between SRAM and DRAM". The small voltage from the memory cell is bumped up by the Sense Amplifier.
       Figure 13-03z7 RAM Addressing	Figure 13-03z7 shows the addressing and read/write schemes for SDAM and DRAM. Multiplexing/De-multiplexing refers to break up and then recombine back the whole for expediting the transmission process. Also see "DRAM Read/Write".

     • Secondary Memory - See Non-Volatile Storages
       
       
   
   
   
   • Operation - The Central Processing Unit (CPU) is the "Heart" of the computer. A summery of its structure (+ hardware implementations) and action (in red color) is shown in Figure 13-03z8, which also depicts an example on instruction steps (for the ADD opcode). Figure 13-03z8a lists the various control signals to run the processes as disigned. The followings start with a brief description of the clock cycle (and why it is necessary), then a little bit more information on each step :
     
     

     Figure 13-03z8 CPU Structure, Data Formats [view large image],	Datapath for R-type (ADD), and I, J Types

     Register file: Small number of internal registers that let multiple values to be read and written simultaneously (for R-type).
     Data memory: Larger grid of memory cells that are used to store main information to be processed by the CPU (for I-type).
     

     Control Unit (CU): It takes in the opcode from the current instruction, and sends signal to the the corresiponding processor. Different signals are generated on each clock cycle to make the overall operation happen. Figure 13-03za links all kinds of CU opcodes to various processors. While Figure 13-03zb lists the values of different CU opcode for various operations (the JtoPC = branch, X = ignored, ALUcontrol is not opcode - it is added to show its relationship with the ALUOp).

     Figure 13-03z8a Control Unit (CU) Signals for ADD [view large image]                 (Figure 13-03z8b)

     • Clock Cycle -
       
       
       • It is possible to build a CPU with continous supply of electrical energy provided there is a mechanism to ensure each step has all the ingredients (from the previous step) to run the processing. This is called asynchronous operation, which requires lot of efforts to maintain. For each subsequent step, the data will not be ready if starting too soon, while the required information would disappear if waiting for too long (see Figure 13-03z9,c).
       
       

       In CPU running with synchronous pulses of electrical energy (1 pulse = 1 clock cycle), the processing of each step has to complete during the interval set by the "machine cycle = n clock cycles" (n 1). Some cycle may have some free time since the temporal requirement would be different for each step and the machine cycle has to be the longest to accommodate all steps (hence the base for "Instruction Pipelining"). Such processing is automatic and doesn't need extra efforts to supervise (see Figure 13-03z9,a and b which also show the Instruction Cycle = N machine cycles for N 1 from beginning to end of an instruction).

       Figure 13-03z9 Clock Cycle [view large image]

       The clock cycle is generated by piezoelectric crystal such as the quartz which behaves like the RLC circuit with less energy loss and very stable. The period can be 10-6 to 10-9 sec depending on CPU model.

       
       
     • Fetch - This step fetches an instruction from memory and stores it in the instruction register (Figures 13-03z8,10,13).

       		Figure 13-03z11 shows the clock cycles consumed in the process and reveals that very little time spends on the bus (omnibus in Latin), It is the retrieval which takes up most of the time - 2 cycles in this example, while there could be more. Meanwhile, this step also increases the Program Counter (PC) by 4 bytes using the ADD instruction to point to the next sequential instruction. Sometimes the PC will be updated with a different value from the execution step (via the control signals such as
       Figure 13-03z10 Fetch	Figure 13-03z11 Clock Cycles for Fetch [view large image]	"Shift left 2 bytes" (for Sign-Extension), "Mux" (2n selector, n = 1, ..., i.e., which one to pick), or "PCsrc" (update PC by 4 bytes). See picture in Figure 13-03z8 and "Datapath for fetch").

     
     
     • Decode - For the ADD instruction (as an example), it reads the values of two source registers specified by the rs and rt

       fields and store these values in the internal registers 1 and 2, respectively in the register file; it would also appoint an output register. The Control Unit decodes (translates the machine codes in the instruction) and generates control signals to various processors. They determines the type of action to be taken (see Figure 13-03z12,z8,z8a,z8b). There is also a "Sign Extend (SE)" field to extend the target address to two 16 bits for J-type instruction. The ALUOut contains the branch target address if the instruction is Jump. The Decode step often piggybacks at the ID/EX boundary between 2 clock cycles.

       Figure 13-03z12 Decode Step [view large image]

       (see Figure 13-03z8 and the image below)

     
     
     • Execution - Depending on the format of the instruction :
       

       	For R-type instructions - this step performs the arithmetic operation on the values contained in the internal registers 1 and 2. The result is placed in the internal ALUOut register (see Figure 13-03z8). For load or store I-type instructions - it follows the steps : R1(addr) + mem(addr/SE) + ALUOp ALU address (in Data Memory) Read Data ALUOut MemtoReg=0 (see Figure 13-03z8).
       Figure 13-03z13 Machine Codes [view large image]	For J-type instructions - the unconditional Jump instruction left-shifts by 2 bytes the 26 least significant bits in the instruction register and adds this amount (2 bytes) to the PC (see Figure 13-03z8).

       • For branch instructions - this conditional J-type instruction compares the values in registers 1 and 2, then assigns the value of the ALUOut register to the PC only if the two values are equal (see Figure 13-03z8).
       
       

       Memory Access - R-type instructions skip this step. For load instructions - Read Data (the address in Data Memory) ALUOut MDR, i.e., the internal Memory Data Register (see Figure 13-03z8). For store instructions - Write Data (from the internal register 2) ALUOut (see Figure 13-03z8). Write Back - This step applies to R-type format only. It writes result back to the ALUOut internal register file, and updates PC (see Figure 13-03z8).

       Figure 13-03z14 Ex...

   • Languages - As mentioned earlier on Control Unit, which translates the machine codes to something that the hardware can

     understand and generating corresponding response. The next level is the Assembly language which translates the machine code to some kind of syntax for comprehension by human (mainly Assembler programmer). Figure 13-03z15 shows an example of "ADD" instruction translating from machine code to Assembler. These are referred to as low-level languages. Over the years, lot of high-level languages appear making coding computer program much easier (but still requires a lot of learning to become proficient, see a short list).

     Figure 13-03z15 Machine-Assembler

     Here's a longer list by programming types and some comments for a few of them :
     
     
     1. Interpreted Programming Languages - Basically, the source codes of such languages are translated into machine codes either at runtime or pre-compiled already. For example in BASIC language, the ADD instruction is just :
        C=A+B,    (where A, B, and C are storages keeping the data, e.g., 1 in A, 2 in B, and 3 in C)
        the programmer doesn't need to worry how the computer does its job (see a DIY example, and BASIC 256 syntax).
        
        Tow more examples are introduced below :
     
     JAVA - It is a pre-compiled programming language depending heavily on Object-Oriented Programming (OOP).

     		Java codes can run on all platforms that support Java without the need to recompile (see Figure 13-03z16 for an example). Python - It uses variety of programming styles and associated language constructs to generate the most suitable programming tool. Figure 13-03z16 is an example of Python program.
     Figure 13-03z16 OOP	Figure 13-03z17 Python Program [view large image]

     2. Compiled Programming Languages - This type requires the programmer to write source codes (the program), which has to be complied and link-edited (with other routines) into an executable object file. The cycle has to repeat even for a slightly changed statement in the source. The FORTRAN (Formula Translation) is the world's first high-level programming language developed at IBM in 1957. It was (and still "is") the programming language of choice for numerical calculations cranking out a lot of PhD theses among numerous applications in science and engineering.
        
        

        	Here's the FORTRAN language reference, and coding sample from the bygone era of FORTRAN IV (circa 1961) :

        The Job Control cards (JCL) to invoke system functions all start at the 1st column with :
        // following by a blank or a "label + a blank",
        //* is for comment,
        the symbol * indicates the dataset is in the following,
        and the symbol & or && signifies a temporary library or dataste (see example in Figure 13-03z18,a).
        
        While the coding rules for source codes on the 80-column card are :

        	Col. 1 : Blank, or a "c" or "*" or ! for comments. Col. 2-5 : Statement label (optional). Col. 6 : Continuation of previous line (optional). Col. 7-72 : Statements. Col. 73-80: Sequence number (optional, rarely used today) Symbol * means default format for READ / WRITE (Figure 13-03z18,b).
        Figure 13-03z18 FORTRAN JCL and Source Program [view large image]	Figure 13-03z18,b is an example to calculate the area A of triangle by A = [s(s-a)(s-b)(s-c)]1/2 where s = (a+b+c)/2

     3. Markup Language - Instead of coding for arithmetic operation, programming has evolved to present textual and graphical information as well in the digital age.
        

        For example, all the webpages in this site are coded in HyperText Markup Language (HTML). Figure 13-03z19 is a very small segment taken from the page on "Monte Carlo Methods". It shows the enrichment of the text by linking to additional references, and a picture to attract attention. Beside adding more information, the presentation in the web is very important for advertiser to promote its products. There are HTML editor for free such as CoffeeCup. The HTML document can be uploaded to the Web Hosting Provider, which will manage the web browsing (for a fee).

        Figure 13-03z19 HTML as Markup Language [view large image]

        The source codes in Figure 13-03z19,a : "The Monte Carlo ....(Figure 01)".

   • Peripherals - The era of personal computer (PC) started in the 1960s with the appearance of desktop computers. It is rather bulky with many parts such as the CPU, memories, lot of slots (PCI, AGP), cards (Video, Sound, Network) packing inside a case plus some external (input/output) devices to interface with the users. They are collectively called Peripherals as described briefly below. Over the years with the miniaturizations of transistor and electronic parts, the keyboard and monitor have been incorporated with the rest into something called laptop. Further shrinking in size would do away with the keyboard altogether (replacing with a touch-screen virtual keyboard for input) producing the tablet.

     Finally, the mobile device considered the keyboard as "inconvenient truth" and becomes just an option (for entering a password at least). The visual streaming and sound (of music) are now the main interest instead of arithmetic computation. Figure 13-03z20 illustrates the evolution. Actually, the desktop is alive and used for serious work. Curiously, gamers prefer desktop because of its superior computing power and storage capacity.

     Figure 13-03z20 Types of Computers [view large image]

     The PC Chipset in Figure 13-03z21 is a typical example of the desktop innards which consist the following parts : Memory slots - for adding RAM to the computer. AGP slot - The slot is for inserting a video card to accelerate 3D computer graphics. PCI slots - The slots can accommodate a variety of cards such as those for video, sound, network, modem, etc. Controllers - for disk drive, networking, USB, etc.

     Figure 13-03z21 CHIPSET

     The rest of the components (not shown) include : clock, buses, and the Basic Input/Output System (BIOS).

   	The external Peripherals include : input devices such as Keyboard, Mouse; output devices such as Monitor, Speakers, Printer; those perform both input, output functions : CD-ROM Drive, Webcam, Scanner, Serial Port Device (such as the Point Of Sale Cash Register), Internet Gateway Box (with modem + router) and USB. See Figure 13-03z22 for some of the Peripherals.
   Figure 13-03z22 Peripherals [view large image]

   
   
5. Operating System (OS) - An operating system is a system software that manages computer hardware, system software, and provides common services for the user's application programs.
   

   The operating system (OS) manages all of the software and hardware on the computer. After initially loaded into the computer by a boot program, it manages all activities in the computer. The application programs make use of the operating system by sending requests for services through the Application Program Interface (API) which connects computers or pieces of software to each other. Users can also interact directly with the operating system through the User Interface (UI), such as the Command-Line Interface (CLI) or a Graphical UI (GUI). Device driver is the system software for hardware interface. The OS brings powerful benefits to user applications without which, every program would need to include its own UI and other interfaces.

   Figure 13-03z23 OS

   Figure 13-03z23 shows the processing of the computer by the user with the OS in between the softwares and hardwares (see "Understanding Operating Systems" for more details).

   The Kernel is the software written for the Operation System, followings are brief description of its various functions :
   
   

   	Program Execution - Executing an application program typically involves the operating system kernel, which assigns memory space and other resources, establishes a priority for the process in multi-tasking systems, loads program binary code into memory, and initiates execution of the application program, which then interacts with the hardware devices and back to the user.
   Figure 13-03z24 OS Processing	An application program can incur (instantiate) many independent processes, e.g., each one of the display in the multiple screens is one standalone process (see Figure 13-03z24).

   • Interrupts - Interrupt occurs not only when the user's application program is abend (abnormal-end, e.g., a number/0 ).

     		It is normal for the OS to control the processing of the various parts in the computer. The OS is similar to the conductor (in orchestra), who directs the tempo of various musical instruments (the hardwares) playing the musical notes (the application programming codes) by the musicians (the user, see Figure 13-03z25). Similarly, interruption of playing certain instrument occur as required by the compositions (the software).
     Figure 13-03z25 Orchestra	Figure 13-03z26 OS Interrupts	The OS interrupt is invoked by :

     • Hardware Interrupt - The CPU would manage the overall processing accordingly upon receiving the interrupt signal from the hardware. For example, the CPU would first initiates the data transfer, then it does other operations while the transfer (with a much slower rate) is in progress and resume the processing (related to that hardware) at the completion of the transfer.
     
     
     • Software Interrupt - It is a type of interrupt caused by an instruction in the program. For example, the processing is waiting for the result from a subroutine. There is an Interrupt handler (called the Interrupt Service Routine, ISR), which is a block of code associated with a specific interrupt condition. After handling the interrupt, the usual programs are executed again if no exception incurred (see some specific examples of "Error Handling").
     
     
     • Processor Interrupt - It is also known as System Interrupt. It provides the vital service to assign clock cycles to

       	various hardware and software processings optimally. Figure 13-03z27 shows its CPU usage of about 3% during a busy time (regular usage would be ~ 0.1%, and 0% for no activity) in the Task Manager display.
       Figure 13-03z27 Task Manager	invoked by ctrl+shift+esc (for Windows 10).

     • Interrupt-driven I/O - This kind of interrupt occurs when the CPU is informed by the hardware that I/O activity is impending. The service is then provided to complete the task without further ado making the process more efficient. The method is especially useful for intermittent movement of keystroke or mouse.
     
     
     • Modes - The OS protects the vital resources by limiting their access only in Supervisor Mode. User applications (in User Mode) are usually forbidden to jump over the barriers except via permission as shown in Figure 13-03z28 which also shows the layers of privileges. Privileged code is managed code that accesses secured resources or performs

       	other security sensitive operations. For example, some device drivers are programs written in highly privileged code. Unprivileged codes are the instructions that are only executed in user mode. Typically, the transfer of control to the supervisor mode is achieved by executing a software interrupt instruction.
       Figure 13-03z28 Modes of Operation	For systems without MMU or MPU, all codes run in supervisor mode, MMU - Memory Management Unit, MPU - Memory Protection Unit.

     
     
   • Memory Management (MM) - It is the process of controlling and coordinating the use of the limited primary memory such

     as the Level 1 cache. The task of memory management is to provide ways (by interrupts) to dynamically allocate portions of the primary memory to programs at their request, and free it for other programs when no longer needed. Such process involves the swapping segment of primary allocation in and out of the disk (virtual) memory (see Figure 13-03z29). Paging is similar but in equal block size (see "Difference between Swapping and Paging"). Page fault is the required segment not found, and to be resolved by MM.

     Figure 13-03z29 Momory Management

     For I/O device with slow rate of data transfer (to the primary memory) such as from the hard-disk, MM delegates the work to a channel or a direct memory access controller; an interrupt is activated only when the transfer is completed.

     
     
   • Multitasking - It allows more efficient use of the computer hardware. For example, when a program is waiting for some external event such as user input or input/output transfer with a peripheral to complete, the idling time can be used to run another program. The OS performs such task using the same methods of interrupt, and swapping/paging.

     	All operating systems today can perform multitasking. When the computer opens the Web browser, Microsoft Word and email at the same time, the operating system is doing multitasking (see Figure 13-03z30). The user is not aware of the switching because the computer's very fast processing rate is beyond comprehension of the human mind.
     Figure 13-03z30 Multitasking

   • Computer File - Computers store data on digital storages with the elementary unit called file. There are organizational structures (hierarchy) in between the OS and file to expedite the work flow (see Figure 13-03z31,a). Here's more details :
     
     
     • Disk Allocation - File space allocation is the method by which data is apportioned physical storage space in the operating system. There are three main disk space or file allocation methods :
       1. Contiguous Allocation - the file occupies space continuously in the form of b, b+1, ... b+(n-1), where b is the size of block (in bytes), n the number of blocks.
       2. Linked - the allocation needs not be contiguous, but required a pointer to the starting and the ending block.
       3. Indexed - it needs a special Index Block contains the pointers to all the blocks occupied by a file.
       
       
     • File System - The file system is the link between the the data physically allocated on the disk and some verbal information in the directory (aka folder) for the end users to manipulate their data. For example, C:\Windows> signifies the allocation of a bunch of file folders for Windows on disk C. In term of the file cabinet analogy, it would be the file folders within one of the drawers (see Figure 13-03z31,a and b).
     
     
     • File Directory - The directory (DIR) contains information about the files, including names, types, sizes, dates/times, attributes,... Much of this information, especially that is concerned with storage, is managed by the operating system. The directory is itself a file, accessible by various file management routines and by the end users. Figure 13-03z31,c is a directory listing from Microsoft's command prompt (cmd.exe), Figure 13-03z31,d is taken from Windows file list.
       To invoke cmd.exc : Task Manager (from File Directory) File Run new task cmd.exe.
       
       Here's the response from ChatGPT (login/signup to ChatGPT) for the question :
       What kind of information is in the file directory?
       
       A file directory, also known as a folder, contains information about the files and other folders that are stored within it. This information typically includes:
       
       

       1. File names: The names of the files stored in the directory. 2. File types: The type of file, such as text, image, audio, video, or executable. 3. File sizes: The size of each file in the directory, typically measured in bytes or kilobytes. 4. File creation and modification dates: The dates and times when each file was created or last modified. 5. File permissions: The access permissions for each file, which determine who can read, write, or execute the file. 6. File attributes: Additional metadata associated with each file, such as the author, keywords, or description. 7. Subdirectories: Information about any subdirectories that are contained within the directory.

       Figure 13-03z31 File System and Directory [view large image]

       File directories are used by computer operating systems to organize and manage files on a storage device, such as a hard drive or flash

       drive. The information stored in a file directory allows users to easily navigate and access their files, and allows software applications to locate and manipulate files as needed. (End Quote).
       
       
     Comments : Further inquiry about "astronomical objects within 3 kpc of the Milky Way center" returns response with similar quality (see "How Nature readers are using ChatGPT"). There are also critical opinions (see for example : "ChatGPT Is Dumber Than You Think"). Considering the data within the whole thing are collected from human (i.e., by Common Crawl et al) with an underpinning of neuro-conection (also see "Neural Network"), plus a lot of trainings; it would returns an acceptable response only for sensible question. It is just garbage in, garbage out.
     Anyway, it is rather reassuring that they decline to answer political and financial questions such as : the chance of Donald Trump to become the US president in 2024 or the buy/sell recommendation of a certain stock. It tries to avoid involvement with those highly unpredictable events but regarded by some as incompetent. As for composing poem, it is just a collection of bland words, i.e., the piece is descriptive (without rhyme and pitch) rather than emotionally stirring (for example, it would not equate "sunset" with the feeling of dying old people).
     
     N.B. The difference between search engine and ChatGPT is the way they present the information. While the search engine gives a list of web sites from which relevant information have to be compiled from each one; ChatGPT would have it done at one fell swoop saving lot of work and avoiding the malicious sites.
     
     By mid March 2023, ChatGPT has rolled out the GPT-4.0 upgrade @ USD $20/mo over the "free" GPT-3.5 (see "GPT-4 is here: what scientists think", "10 ways GPT-4 is impressive but still flawed" and "Comparison of Performance for Various Subjects").
     
     

     There are lot of discussions about AI subjugating or exterminating the human race (see "How Frightened Should We Be of AI ?"). Such scenario will never happen as all living organisms generate their energy internally through either photo-synthesis or by consuming carbohydrate (eventually turning into numerous tiny organic batteries called ATP, see "Why ATP ?"); while the AIs are machines powered by electricity which is controlled by human. The image on the left illustrates the

     termination of a scheming AI by a human. It is from the trailer of the 1968 movies "2001: A Space Odyssey", see : details.

     End of Commets
     
     

     	File - A file is a collection of related information that is recorded on storage. The related subject is indicated by the file type such as text, image, video, ... (see a list in Figure 13-03z32). From the user's perspective, a file is the smallest allotment of data storage. All I/O are done at this level. The "File Format Overview and Information" lists a huge amount of different
     Figure 13-03z32 Filetypes	file types available. To get more detailed information about each file extension, just click on the links or the " > ". It also previde examples.

     
     
   • Networking - Network Operating System (NOS) is an Operating System with Networking capability, i.e., software.
     
     One of the 2 types allows Personal Computer (PC) to participate in a client-server system in which the server enables multiple clients (PCs) to share its resources, such as printer, storage, web and mail, ... Figure 13-03z33,a shows the client-server configuration including a central Hub (comprised of modern and router) which directs traffic in between the PCs and servers with wire or wireless.
     
     The other type of NOS is the "peer to peer" Network shown in Figure 13-03z33,b. It is a group of computers that share each

     other's resources within a Local Area Network (LAN). Every one in the network keeps track of its own user accounts and security settings and no single computer is in charge of the workgroup. The same image also adds the processing in the internet environment. It involves an additional layer of Internet Service Provider (ISP) for the exchange of information between the users and whatever in the internet. The users pay the ISP for such service.

     Figure 13-03z33 NOS [view large image]

     Anyway, all participating members in whatever networking scheme would have NOS installed in their device for such incredible event to happen.

   • Types of OS - there are several types of OS installed on different types of computers (see Figure 13-03z34) :
     
     
     • Main Frame Computer - It used to be dominated by IBM/360 in the mid-1990s. It has incarnated through a series of new versions, the latest one in 2021 is z/OS version 2. The markets include the world's financial markets and much of global commerce. It runs user applications such TSO, CLIST, ... never heard of by the general public.
     
     
     • Windows - It is very popular with PC users. Windows 10 has a very reliable OS and security from personal experience.
     
     

     	Android - Android is a mobile operating system developed by Google for touchscreen devices, cell phones, and tablets. It is the most popular OS with 40% market share worldwide.
     Figure 13-03z34 OS Types [view large image]	iOS - It is a mobile operating system developed by Apple Inc. exclusively for its hardware with about 16% market share worldwide among the mobile phone market.

     
     
   • User Interface - It either uses the "Command Line Interface" by typing out the input line-by-line in MS-DOS PowerShell such as the one shown in Figure 13-03z35.
     The procedure :
     
     • Invoke the PowerShell from Windows PowerShell in file directory.

     		Check out the available commands from "List of DOS commands". Enter the command such as "date". Get date. Go to next one. Enter "exit" to get out.
     Figure 13-03z35 DOS Command [view large image]	Figure 13-03z36 Graphical Interface	Or click an icon by using the "Graphical Interface" such as the "Go!" in Figure 13-03z36 to get date. The OS would do the rest.

ChatGPT said:
A Graphics Processing Unit (GPU) is a highly parallel processor originally designed to accelerate the rendering of images and video. Over time, its role has expanded far beyond graphics. Its main functions include:
________________________________________
1. Graphics Rendering
" Rasterization: Converting vector graphics (shapes, 3D models) into pixels on a screen.
" Shading: Applying colors, textures, lighting, and shadows to make objects realistic.
" Transformations: Rotating, scaling, and projecting 3D objects into 2D display coordinates.
" Frame Buffering: Storing image data (frames) before displaying them on the monitor.
________________________________________
2. Parallel Computation
" GPUs contain thousands of smaller cores optimized for performing many operations simultaneously.
" This makes them ideal for parallelizable workloads, unlike CPUs, which are optimized for sequential tasks.
________________________________________
3. General-Purpose GPU Computing (GPGPU)
" Using APIs like CUDA (NVIDIA) or OpenCL, GPUs can be used for non-graphics tasks such as:
o Scientific simulations (physics, chemistry, climate models)
o Machine learning & AI (training deep neural networks)
o Cryptocurrency mining (solving hashing problems)
o Medical imaging (CT/MRI data processing)
o Data analytics & big data processing
________________________________________
4. Video Processing
" Decoding/encoding video streams (e.g., H.264, H.265, AV1).
" Post-processing: Scaling, filtering, and improving video playback quality.
" Hardware acceleration: Offloading video tasks from CPU to save power and improve performance.
________________________________________
5. Real-Time Interaction
" Powering 3D games, VR/AR applications, and simulations that require smooth frame rates.
" Handling physics simulations (rigid body dynamics, fluid motion) in real-time rendering engines.
________________________________________
6. AI & Machine Learning Acceleration
" Modern GPUs have Tensor Cores or similar units specifically optimized for matrix operations, vital for neural networks.
" Used in both training and inference for AI applications.
________________________________________
7. High-Performance Computing (HPC)
" In supercomputers, GPUs work alongside CPUs to achieve massive performance in research and engineering applications.
" They are central to exascale computing efforts.
________________________________________

? In summary:
The GPU's primary role is still graphics rendering, but it has evolved into a parallel computing powerhouse used in AI, science, engineering, and media processing due to its ability to execute many operations at once.

Would you like me to also explain how the GPU differs from the CPU in its function and architecture?

ChatGPT said:
Great question ?? - understanding how a GPU differs from a CPU explains why they're used for different tasks.
________________________________________
??? CPU vs. GPU: Function & Architecture
1. Core Design
" CPU (Central Processing Unit):
o Few cores (typically 4-32 in consumer systems, more in servers).
o Each core is powerful and optimized for sequential tasks.
o Handles complex logic, branching, and varied workloads.
" GPU (Graphics Processing Unit):
o Thousands of smaller, simpler cores.
o Designed for parallel execution of the same operation on many data points.
o Ideal for workloads like graphics rendering, matrix multiplications, and simulations.
________________________________________
2. Task Specialization
" CPU:
o General-purpose processor, the "brain" of the computer.
o Best at tasks that require low latency and decision-making (OS control, running apps, logic-heavy code).
" GPU:
o Specialized co-processor for high-throughput computation.
o Best at tasks with massive parallelism (rendering pixels, AI training, cryptography, etc.).
________________________________________
3. Instruction Handling
" CPU:
o Optimized for single-thread performance.
o Handles complex instructions, branching, and unpredictable workloads.
" GPU:
o Optimized for SIMD (Single Instruction, Multiple Data) - running the same instruction on many data simultaneously.
o Limited for general-purpose logic-heavy code, but excels at uniform math tasks.
________________________________________
4. Memory Architecture
" CPU:
o Larger cache (L1, L2, L3), sophisticated memory hierarchy.
o Optimized for fast access to small amounts of data.
" GPU:
o High-bandwidth memory (GDDR6, HBM) for moving large data sets quickly.
o Smaller cache per core, but relies on throughput over latency.
________________________________________
5. Use Cases
" CPU:
o Running the operating system, browsers, word processors.
o Managing system I/O, decision-making, and sequential logic.
" GPU:
o Rendering 3D graphics in games and movies.
o Training AI models with billions of parameters.
o Performing repetitive mathematical calculations on large data sets.
________________________________________
?? Analogy
" CPU = A few skilled workers who can solve complex problems one at a time, very efficiently.
" GPU = A factory with thousands of workers who can all do the same simple job at once, extremely quickly.
________________________________________
?? In practice:
" For general computing, you need a CPU.
" For parallel heavy lifting (graphics, AI, simulations), you need a GPU.
" Modern systems combine both - the CPU acts as the "manager," the GPU as the "workforce."
________________________________________

Would you like me to also show you a diagram of CPU vs. GPU architecture so it's visually clear?

NVIDIA ()

ChatGPT said:
You're absolutely right-while NVIDIA continues to dominate the GPU market, several competitors offer viable alternatives across different segments. Here's a comprehensive breakdown:
________________________________________
1. Main Competitors in the GPU Market
AMD (Advanced Micro Devices)
" Flagship Alternative: Offers both consumer GPUs (Radeon RX series) and professional/AI accelerators (e.g., Radeon AI Pro R9700) Tom's HardwareThe Verge.
" Market Share: Holds roughly 8% of the discrete GPU (AIB) market as of Q1 2025, down from previous years WccftechTechPowerUpTom's Hardware.
" Known Strengths: Strong mid-range performance, better price-to-performance ratios, and energy efficiency. However, uptake of their latest RDNA 4 lineup remains modest The VergeTechRadar.
Intel (Arc GPUs)
" Discrete GPU Entry: Intel's Arc line (Alchemist, Battlemage, etc.) targets gaming and professional markets Wikipedia.
" Market Share: Now virtually zero in the discrete GPU arena, with past presence evaporating by Q1 2025 WccftechPCGamesNTom's HardwareTechPowerUp.
" Potential Levers: Integrated ecosystem and XeSS upscaling tech (similar to DLSS/FSR) could give it a foothold if software and hardware align well TechnoSports Media Group.
________________________________________
2. Specialized and Emerging Players
Google (TPU)
" Not a GPU per se, but widely used for AI workloads within Google's infrastructure, providing an alternative to Nvidia's
AI chips TechRadar.
AI Chip Startups & Innovators
" Companies like Cerebras, Groq, SambaNova, Etched, and Qualcomm are winning traction in niche use cases-e.g., high-frequency trading, recommendation systems, sovereign cloud deployments-by offering highly customized or efficient AI accelerators Business Insider.
________________________________________
3. Regional & Emerging GPU Vendors (China Focus)
Jingjia Micro (Changsha Jingjia)
" One of China's most notable GPU developers with independent IP and production capability.
" Offers GPUs like the JM9 series, comparable in performance to older Nvidia models (e.g., GTX 1080) Wikipedia. Cambricon Technologies
" Develops GPGPU and AI-focused chips, often compared to Nvidia due to similar AI aspirations.
" Recently benefited from China's push for domestic alternatives and reached profitability Wikipedia.
MetaX Integrated Circuits
" A newer Shanghai-based company that designs GPGPU chips for AI workloads.
" Founded by ex-AMD engineers and actively preparing for public listing (IPO), backed by both government and private funds Wikipedia.
________________________________________
Market Snapshot: Who's in the Game?
Segment Key Competitors Notable Strengths / Focus
Discrete Desktop GPUs NVIDIA (dominant), AMD, Intel (negligible) High performance; gaming & AI capabilities
AI Accelerators Google TPUs, Cerebras, Groq, Etched, SambaNova, Qualcomm Niche-specific high-efficiency compute Domestic/Regional (China) Jingjia Micro, Cambricon, MetaX Sovereign tech with local production and R&D
________________________________________

Key Insights
" NVIDIA remains the clear market leader, with over 92% share in the discrete (AIB) GPU market as of Q1 2025-a high-water mark largely driven by their RTX 50 series launch WccftechTechPowerUpEpiumBusiness Insider.
" AMD continues to compete, especially on price-performance and energy efficiency, though still dwarfed in market share.
" Intel has yet to make a meaningful impact in discrete GPUs, but could leverage its CPU dominance and tech ecosystem for gradual growth.
" Specialized AI chips (TPUs, Cerebras, etc.) are carving out new spaces, particularly in enterprise and AI sectors.
" China's domestic players (Jingjia, Cambricon, MetaX) are accelerating under government support, focusing on local markets and strategic independence.
________________________________________

Would you like a deeper dive into any particular competitor-like AMD's AI roadmap, Intel's Arc future, or what's next from those Chinese startups?

ChatGPT said:
That's a sharp observation. On the surface it seems contradictory, since Xi Jinping and Donald Trump (and U.S.-China policy in general) often clash. Yet NVIDIA has managed to navigate between both governments-at least so far-because of how critical its technology has become.

Here's how it works:
________________________________________
1. NVIDIA's Value to the U.S. (Trump, Washington)
" National Tech Champion: NVIDIA is the leading U.S. chip company, driving AI, gaming, and supercomputing. Supporting it aligns with Trump's "America First" industrial policy, which favors domestic tech winners.
" Stock Market & Economy: NVIDIA's meteoric rise in valuation boosts Wall Street and retirement funds-Trump likes to highlight strong markets as proof of U.S. prosperity.
" Military & Security Use: U.S. defense and intelligence increasingly use GPUs for AI, simulations, and cyber-defense. This makes NVIDIA a "strategic asset" worth backing politically.
" Compliance with Restrictions: When Washington imposed export controls on high-end AI GPUs to China, NVIDIA adapted quickly, designing "China-only" downgraded models (like the A800/H800). This showed NVIDIA's willingness to follow U.S. rules without completely losing the Chinese market.
________________________________________
2. NVIDIA's Value to China (Xi, Beijing)
" Indispensable AI Supplier: Despite China's push for semiconductor independence, NVIDIA's GPUs remain the gold standard for training large AI models. Chinese firms (Baidu, Tencent, Alibaba, ByteDance) rely heavily on them.
" Workaround Chips: The special "China-only" GPU variants (slower interconnect speeds to meet U.S. export rules) still allow Chinese AI labs to keep progressing. For Xi, it's better than being cut off entirely.
" Symbol of Cooperation: Xi can point to NVIDIA's continued presence as proof that China can still access world-class technology, despite U.S. sanctions.
" Strategic Patience: While Beijing invests in homegrown alternatives (Cambricon, MetaX, Biren), keeping ties with NVIDIA buys China time until domestic GPUs catch up.
________________________________________
3. How NVIDIA Balances the Tightrope
" Dual Messaging:
o To Washington: "We protect U.S. IP (Internet Protocol) and follow export laws."
o To Beijing: "We're committed partners and want to help your AI industry grow-within the rules."
" Business Reality: China is ~20-25% of NVIDIA's revenue. Too valuable to abandon, but too politically sensitive to supply unrestrictedly.
" Diplomatic Neutrality: Jensen Huang (NVIDIA's CEO) positions the company as apolitical, focused on innovation and customer demand - never openly siding with one government against the other.
________________________________________
4. Why Both Xi and Trump Approve
" Trump sees NVIDIA as an American winner, compliant with export controls, boosting U.S. markets, and helping national security.
" Xi sees NVIDIA as an indispensable partner, still willing to sell China modified GPUs despite restrictions, ensuring Chinese AI doesn't stall completely.
________________________________________

? In short: NVIDIA manages to please both by being indispensable to each side for different reasons-for the U.S., a national champion; for China, a critical supplier. Its ability to adapt product lines to political constraints lets both Xi and Trump claim a kind of victory.
________________________________________
Would you like me to also explain what risks NVIDIA faces long-term in trying to keep both sides happy? (It could backfire if tensions rise further.)

ChatGPT said:
Here's a side-by-side comparison of parallel processing features in the human brain, shark brain, and a GPU.
" Humans: highly flexible, specialized, and efficient across many tasks.
" Sharks: strong parallelism but more specialized for survival (sensory fusion).
" GPUs: maximum raw parallel number-crunching, but rigid and energy-hungry.
Would you like me to also draw a diagrammatic flow view (like boxes and arrows) to show how information flows in parallel through each system?

ChatGPT said:
Perfect - let's take an evolutionary perspective on parallel processing, from the simplest life to the human brain, and then to GPUs:
________________________________________
1. The Origin of Parallelism in Life
" Even single-celled organisms (bacteria, protozoa) process information in parallel.
o Example: a bacterium senses chemical gradients, temperature, and light, all at once, to decide how to swim.
o Each receptor and signaling pathway is like a "mini-processor," running simultaneously.
" This is the most primitive form of parallel processing: chemical computation.
________________________________________
2. Early Nervous Systems (Jellyfish, Worms)
" Around 600 million years ago, simple animals evolved nerve nets - webs of neurons without a central brain.
o Every part of the body can react simultaneously to stimuli (like a touch or light).
o Still very parallel, but with no central coordination.
" This system resembles a distributed network more than a modern computer.
________________________________________
3. Specialized Brains (Fish, Sharks, Reptiles)
" With predators and prey, animals evolved specialized brain regions for vision, smell, movement.
" Sharks are a prime example:
o Electroreception, smell, and lateral line all processed in parallel, then integrated for hunting.
" This was a huge evolutionary leap: parallel streams feeding into decision-making.
________________________________________
4. Mammals and Humans
" The neocortex added layers of abstraction and planning.
" Humans don't just process multiple senses - we also run parallel mental tasks: planning dinner while walking, listening, and keeping balance.
" This is parallelism with flexibility: the same neurons can switch roles depending on context.
________________________________________
5. GPUs (Our Attempt at Parallelism)
" Inspired by biological parallelism, but simplified:
o Thousands of identical cores doing the same operation at the same time.
o Very powerful for math-heavy tasks (matrix multiplications in AI).
o But no adaptability, no sensory integration, no survival logic - just raw math.
" In evolutionary terms, a GPU is like a giant nerve net, not a brain.
________________________________________
Big Picture
" Life invented parallelism billions of years ago, starting with chemical sensors in microbes.
" Nervous systems layered specialization and coordination on top of it.
" Human brains refined it into flexible, abstract thinking.
" GPUs are our engineering shortcut: taking one trick from nature (parallel connections) and scaling it, but leaving behind most of what makes life intelligent.
________________________________________
?? So yes - GPUs are a kind of imitation of life's parallelism, but compared to evolution's design, they're flat and rigid.

Would you like me to make a timeline-style diagram showing this progression: bacteria ? nerve nets ? shark brain ? human brain ? GPU?

Go to Next Section
 or to Top of Page to Select
 or to Main Menu