<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Open Row</title>
    <description>The latest articles on DEV Community by Open Row (@open_row_3ba1a2dd6799f1c9).</description>
    <link>https://dev.to/open_row_3ba1a2dd6799f1c9</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4258197%2F36ec090b-0673-4f8f-aee9-7c7ef93ef904.png</url>
      <title>DEV Community: Open Row</title>
      <link>https://dev.to/open_row_3ba1a2dd6799f1c9</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/open_row_3ba1a2dd6799f1c9"/>
    <language>en</language>
    <item>
      <title>Why DRAM Reads Are Destructive (and Why That Explains Half of DDR Timing)</title>
      <dc:creator>Open Row</dc:creator>
      <pubDate>Sun, 22 Nov 2026 02:57:22 +0000</pubDate>
      <link>https://dev.to/open_row_3ba1a2dd6799f1c9/why-dram-reads-are-destructive-and-why-that-explains-half-of-ddr-timing-1lc8</link>
      <guid>https://dev.to/open_row_3ba1a2dd6799f1c9/why-dram-reads-are-destructive-and-why-that-explains-half-of-ddr-timing-1lc8</guid>
      <description>&lt;p&gt;"Why are DRAM reads destructive?" shows up in a lot of hardware interviews. The one-line answer is "charge sharing." The answer that survives follow-up questions takes about three minutes, and it explains why tRAS, tWR, and refresh exist at all.&lt;/p&gt;

&lt;h2&gt;
  
  
  One transistor, one capacitor
&lt;/h2&gt;

&lt;p&gt;An SRAM bit is six transistors holding a latch: it actively keeps its value. A DRAM bit is one transistor and one capacitor (1T1C). It just stores charge. That's why DRAM is so dense and cheap per bit, and also why it's strange to read.&lt;/p&gt;

&lt;h2&gt;
  
  
  Reading is charge sharing
&lt;/h2&gt;

&lt;p&gt;Before a read, the bitline is precharged to the midpoint, VDD/2. When the wordline rises, the cell's small capacitor connects to the bitline's much larger capacitance, and they settle to one shared voltage:&lt;/p&gt;

&lt;p&gt;ΔV = (V_cell − VDD/2) × Cs / (Cs + C_BL)&lt;/p&gt;

&lt;p&gt;With illustrative values (VDD = 1.1 V, Cs = 10 fF, C_BL = 40 fF), a stored 1 moves the bitline by 0.55 × 10/50 = &lt;strong&gt;110 mV&lt;/strong&gt;. A tiny signal.&lt;/p&gt;

&lt;p&gt;Now look at the cell. Its capacitor is sitting at the shared voltage, not at a full 1. Reading it mostly erased it.&lt;/p&gt;

&lt;h2&gt;
  
  
  The sense amplifier reads &lt;em&gt;and&lt;/em&gt; restores
&lt;/h2&gt;

&lt;p&gt;A cross-coupled sense amplifier sees that 110 mV, regenerates it with positive feedback to full rail, and, because the access transistor is still on, drives the full value back into the cell. One activate does this for an entire row at once; that row of sense amps is the &lt;strong&gt;row buffer&lt;/strong&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why this explains the timing rules
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;tRCD&lt;/strong&gt;: you can't read columns until the sense amps have resolved.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;tRAS&lt;/strong&gt;: you can't precharge until the restore into the cells is finished.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;tWR&lt;/strong&gt;: written data must reach the capacitors before the row closes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Refresh&lt;/strong&gt;: capacitors leak anyway, so every row is periodically activated just to restore it.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;An interviewer follow-up you should expect: &lt;em&gt;"Why can't a bank have two rows open?"&lt;/em&gt; Because a bank has one set of bitlines and sense amps. A second open row would charge-share onto the same bitlines and corrupt both.&lt;/p&gt;

&lt;h2&gt;
  
  
  Keep going
&lt;/h2&gt;

&lt;p&gt;I wrote a free one-page DDR5 timing cheat sheet that maps each parameter to its physical cause. &lt;a href="https://openrow.gumroad.com/l/ddr5-cheat-sheet" rel="noopener noreferrer"&gt;Download it free&lt;/a&gt;. If you want the full treatment, with worked DDR5-4800 examples and 30 interview questions with model answers, there's the &lt;a href="https://openrow.gumroad.com/l/ddr5" rel="noopener noreferrer"&gt;DRAM &amp;amp; DDR5 Interview Field Guide&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>hardware</category>
      <category>interview</category>
      <category>beginners</category>
      <category>computerscience</category>
    </item>
    <item>
      <title>DDR5 Timing Parameters Explained by Physics (tRCD, tRAS, tRP, tRRD, tFAW)</title>
      <dc:creator>Open Row</dc:creator>
      <pubDate>Sun, 22 Nov 2026 02:56:47 +0000</pubDate>
      <link>https://dev.to/open_row_3ba1a2dd6799f1c9/ddr5-timing-parameters-explained-by-physics-trcd-tras-trp-trrd-tfaw-429</link>
      <guid>https://dev.to/open_row_3ba1a2dd6799f1c9/ddr5-timing-parameters-explained-by-physics-trcd-tras-trp-trrd-tfaw-429</guid>
      <description>&lt;p&gt;If you're preparing for an RTL, verification, or memory interview, you've probably memorized what tRCD stands for. The follow-up question is the hard one: &lt;em&gt;what is the silicon actually doing during that wait?&lt;/em&gt; Once you see that every timing parameter is analog physics with a digital name, you can rebuild most of the rules from scratch.&lt;/p&gt;

&lt;h2&gt;
  
  
  Start with the cell
&lt;/h2&gt;

&lt;p&gt;A DRAM bit is charge on a tiny capacitor, connected to a bitline through one access transistor. Reading it works by &lt;strong&gt;charge sharing&lt;/strong&gt;: the bitline is precharged to VDD/2, the wordline turns on, and the small cell capacitor nudges the much larger bitline up or down by roughly 100 mV. A sense amplifier then amplifies that difference to full rail, and in doing so writes the value back into the cell, because charge sharing destroyed it.&lt;/p&gt;

&lt;p&gt;Every core timing parameter protects one step of that process.&lt;/p&gt;

&lt;h2&gt;
  
  
  The parameters, one physical cause each
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Parameter&lt;/th&gt;
&lt;th&gt;Rule&lt;/th&gt;
&lt;th&gt;What the silicon is doing&lt;/th&gt;
&lt;th&gt;DDR5-4800 typical&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;tRCD&lt;/td&gt;
&lt;td&gt;ACT → RD/WR&lt;/td&gt;
&lt;td&gt;Charge sharing plus sense-amp latching&lt;/td&gt;
&lt;td&gt;16 ns&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;tRAS&lt;/td&gt;
&lt;td&gt;ACT → PRE&lt;/td&gt;
&lt;td&gt;Sense amp finishing the restore into the cells&lt;/td&gt;
&lt;td&gt;32 ns&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;tRP&lt;/td&gt;
&lt;td&gt;PRE → ACT&lt;/td&gt;
&lt;td&gt;Bitlines equalizing back to VDD/2&lt;/td&gt;
&lt;td&gt;16 ns&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;tRC&lt;/td&gt;
&lt;td&gt;ACT → ACT, same bank&lt;/td&gt;
&lt;td&gt;One full open-restore-close cycle (tRAS + tRP)&lt;/td&gt;
&lt;td&gt;48 ns&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;tWR&lt;/td&gt;
&lt;td&gt;Last write data → PRE&lt;/td&gt;
&lt;td&gt;New data reaching the cell capacitors&lt;/td&gt;
&lt;td&gt;30 ns&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;tRRD_S / _L&lt;/td&gt;
&lt;td&gt;ACT → ACT, other bank&lt;/td&gt;
&lt;td&gt;Activate current spikes: protecting the power grid&lt;/td&gt;
&lt;td&gt;8 nCK / max(8 nCK, 5 ns)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;tFAW&lt;/td&gt;
&lt;td&gt;≤ 4 ACTs per window&lt;/td&gt;
&lt;td&gt;Same power limit over a longer window&lt;/td&gt;
&lt;td&gt;device-dependent&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;tCCD_S / _L&lt;/td&gt;
&lt;td&gt;Column → column&lt;/td&gt;
&lt;td&gt;Bank groups share internal data paths&lt;/td&gt;
&lt;td&gt;8 nCK / max(8 nCK, 5 ns)&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Two of these trip people up. &lt;strong&gt;tFAW and tRRD are not logic delays.&lt;/strong&gt; Every activate fires a whole row of sense amplifiers at once, a big current spike. Too many in a short window and the internal supply droops. &lt;strong&gt;The _S/_L split exists because of bank groups:&lt;/strong&gt; banks in the same group share prefetch circuitry, so same-group operations need more spacing.&lt;/p&gt;

&lt;h2&gt;
  
  
  The classic whiteboard question: tRRD_L in clocks
&lt;/h2&gt;

&lt;p&gt;DDR5-4800 means 4800 MT/s. Data moves on both clock edges, so the clock is 2400 MHz and tCK ≈ 0.4167 ns.&lt;/p&gt;

&lt;p&gt;The rule is max(8 nCK, 5 ns). 5 / 0.4167 = 12, so &lt;strong&gt;tRRD_L = 12 clocks&lt;/strong&gt;, not 8. Hard-coding the "8 nCK" you see in a table is an easy bug that every test written from the same assumption will happily pass.&lt;/p&gt;

&lt;h2&gt;
  
  
  Want the one-page version?
&lt;/h2&gt;

&lt;p&gt;I put every core DDR5-4800 parameter on a free printable cheat sheet: the rule, the one-line physics, the typical value, and the clock count. &lt;a href="https://openrow.gumroad.com/l/ddr5-cheat-sheet" rel="noopener noreferrer"&gt;Grab it free here&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Values are typical published figures for DDR5-4800; always check your device datasheet and JEDEC JESD79-5. Drafted with AI assistance and reviewed for accuracy.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>hardware</category>
      <category>career</category>
      <category>interview</category>
      <category>beginners</category>
    </item>
  </channel>
</rss>
